Recent evidence shows that subtle impairments in visual processing characterise some individuals with dyslexia. Specifically, impaired processing is found for low spatial and high temporal frequency visual stimuli (Lovegrove et al., 1980; Martin & Lovegrove, 1987; Talcott et al., 1998; but see also Skottun, 2000; Walther-Muller, 1995). The most robust of these findings indicates a particular problem with visual motion processing (Cornelissen et al., 1994; Demb et al., 1998; Talcott, Hansen, Elikem & Stein, 2000; Everatt, Bradshaw & Hibbard, 1999). Low spatial frequency, high temporal frequency and motion information are carried primarily via one of two parallel retino-cortical pathways in the human visual system: the magnocellular pathway. Thus, a unifying theory of visual factors in dyslexia holds that there is a magnocellular channel deficit. Neuroanatomic (Galaburda & Livingstone, 1993; Livingstone, Rosen, Drislane & Galaburda, 1991) and fMRI (Eden, van Meter, Rumsey, Maisog & Zeffiro, 1996; Demb, Boynton & Heeger, 1997) evidence consistent with this hypothesis corroborates the perceptual findings. Yet it is still unclear whether these visual deficits represent a correlate or a cause of reading problems. A direct causal link between vision and dyslexia is suggested by findings and observations where reading under demanding visual conditions, such as with small print, leads to declines in reading performance by dyslexic individuals (Cornelissen, Bradley, Fowler & Stein, 1991; Skottun, 2001).

Cornelissen et al. (1991) found that reading errors decreased with larger print size for reading disabled children with poor binocular control, suggesting a causal link between a stressed visual system and reading impairment. Our previous study (O'Brien, Mansfield & Legge, 2000) also implicated a potential effect of print size, where dyslexic readers require larger print sizes than normal readers to achieve their maximum reading speed. The print size effect was observed between subjects where the group difference in reading speed was greatest with small print, but the number of participants with repeated measures data across print sizes was too small to warrant a firm conclusion. Skottun (2001) noted the importance of demonstrating a parametric dependence of dyslexic reading on a visual variable, and pointed to this potential print size effect in our data. Such an effect would support a link between vision and dyslexia because a purely phonological explanation would not predict an effect of print size.

It is possible that larger print size facilitates dyslexic reading by increasing the visibility of spatial frequencies critical for letter recognition (Solomon & Pelli, 1994) or reading (Legge, Pelli, Rubin & Schleske, 1985). In addition, it is possible that dyslexic readers require a wider spatial-frequency bandwidth for reading than the two cycles per letter measured by Legge et al. (1985). For example, with larger letters, more distinguishing letter features may be available within the two cycle/character critical bandwidth for reading. Dyslexic readers may require more or different sets of features to distinguish between letters within words either because they are less efficient at picking up this visual information than skilled readers, or because their memory store of letter templates to match to incoming information is less well established. Other related factors that could be ameliorated with larger print include lateral masking (‘crowding’) effects, that may be greater in dyslexic readers (e.g. Geiger & Lettvin, 1987, but see also Klein et al., 1990), or differences in visual span, that is, the number of letters recognisable in a glance (Legge, Ahn, Klitz & Luebker, 1997; Legge, Mansfield & Chung, 2001).

An alternative account of observed print size effects is that they could result from a general magnocellular deficit, which has been proposed to interfere with letter position coding (Cornelissen et al., 1998a, 1998b). The magnocellular channel provides the primary input to the dorsal occipito-cortical pathway, the so-called ‘where’ pathway that carries information about an object's position in space (Mishkin, Ungerlieder & Macko, 1983). For reading, fidelity of the magnocellular channel should help to localise correctly the serial position of letters within words, and a dysfunction could lead to improper letter localisation and therefore reading errors (Cornelissen & Hansen, 1998). In this case, it may not be the visibility of certain letter features, but instead their relative position within words that is affected by the size of the letters. It could be that deficient position signals have a bigger impact for very small letters whose relative position is harder to identify, and so may result in poorer reading performance with small print.

A magnocellular dysfunction in dyslexic reading has also been related to deficits in visual attention (Iles, Walsh & Richardson, 2000; Pammer & Wheatley, 2001). This theory proposes specific problems with serially focusing attention, either through rapid attentional capture or with attentional disengagement (Valdois, Bosse & Tainturier, 2004). In this regard, Pammer, Lavis, Hansen and Cornelissen (2004) found evidence that problems with attentional focusing contribute to anomalies in relative position coding in dyslexia. Specifically, they found that accurate symbol string recognition was negatively affected by adjacent symbol reversals. Thus, while attentional disturbances could affect reading in their own right, visual attention deficits may manifest as letter position coding errors as well.

To study in detail the trend that we found previously for group differences of print size effects on reading performance, we compared reading rates as a function of print size for groups of children with and without dyslexia. We measured reading rate across a range of twelve print sizes from −0.2 to 1.0 logMAR (corresponding to a range of 20/12 to 20/200 Snellen letter sizes, or 0.037 cm to 0.582 cm in x-height values) for each individual, and fit the reading rate by print size curves with a two-limb fit that has been found to characterise skilled reading (Mansfield, Legge & Bane, 1996). The two-limb fit captures how reading rates are typically constant across a range of large print sizes, then drop off rapidly below a critical print size (for example, see Figure 1).

Although the dyslexic profiles had the normal shape, two characteristics of the two-limb fits did differ for the dyslexic readers. As expected, their maximum reading speed along the plateau was slower than that of the non-dyslexic readers even though they were older and with control for word identification skill. Also, their critical print sizes were larger than those of the controls, indicating that they needed larger print to support maximum reading rates. The group difference in critical print size was partially due to a difference in slopes of the descending portion of the two-limb fits. Dyslexic readers' shallower slopes were moderately correlated with larger critical print sizes (r = −0.58).

These results confirm the hypothesis that dyslexic print size-by-reading rate curves have the typical two-limbed form but larger critical print sizes, and refute the alternative hypothesis of a monotonic increase of dyslexic reading rates with increasing print size. What could theoretically account for these results? Reading words with very small letters may prove difficult either because the letters are hard to identify or because it is hard to determine where they appear in the word. Models of reading hold that both letter identification and letter position coding play important roles in word identification (e.g. Adams, 1979; Estes & Brunn, 1987). Thus, slow, inaccurate word identification with small letters could be due to either compromised letter identification or letter position coding.

Letter recognition and reading normally require a critical spatial frequency bandwidth. The idea that dyslexic readers need larger print to compensate for a higher or broader spatial frequency bandwidth for reading does not account for the findings, since one would expect reading performance to keep improving with increasing print size if this were the case. Likewise, the idea that dyslexic readers may have a particular print size where reading performance peaks, due to differential sensitivity to specific spatial frequencies, was also not borne out. There were no points above critical print size where reading was more efficient for the dyslexic readers.

Letter recognition alone is not sufficient for word identification and reading; rather inter-letter effects also need to be taken into account. Knowing the position of a letter relative to the other letters in a word is necessary for correct word identification; for example, distinguishing the orthographic input of ‘trap’ versus ‘tarp’ or ‘part’ requires correctly locating the spatial arrangement of the letters in the word. The magnocellular channel may carry a code for the relative position of each letter in the word; for instance, [t-1], [r-2], [a-3], [p-4]. This channel may contain more noise in dyslexic readers, so the numbers for ‘r’ and ‘a’ in our example may contain some uncertainty (e.g. for position 2 in the word there is a 60% chance it is the ‘r’ that I saw and a 40% chance it is the ‘a’). Cornelissen and Hansen (1998) describe how relative letter position information is contained within an intermediate spatial scale of the visual stimulus (i.e. the print), while information about whole words and letter features is carried by coarse and fine scales, respectively.

For very tiny print, the fine spatial scale must become too small to resolve critical features, so letters cannot be identified and the acuity limit is reached. Just above this point, letters can be resolved but may not be ordered correctly because of noise (spatial uncertainty) in the codes for spatial position. Theoretically, greater relative position noise in dyslexia may contribute to slow reading in general, and could cause dyslexic reading to be more susceptible to deterioration with small print: meaning dyslexic readers should have slower maximum reading speeds and higher critical print sizes. The present results are consistent with a dyslexic magnocellular-letter position coding deficit.

Other related accounts of inter-letter effects on word reading include crowding, visual attention and visual span. Crowding, or lateral masking, refers to the hindering of letter recognition when a letter is flanked by other letters. The difference in curve shape for dyslexic readers could be because of greater susceptibility to crowding effects, especially if their recognition of isolated letters is already inaccurate or slow. Indeed, previous studies have found greater crowding effects for dyslexic readers (Atkinson, 1999; Geiger & Lettvin, 1987). Klein and D'Entremont (1999) found greater interference between far flankers than near ones for dyslexics, which they attributed to differences in visual attentional filtering. While it is unclear how visual attention differences per se could account for the present print size effect in dyslexia, two theories are relevant. First, visual attention could be the cognitive module that mediates between magnocellular function at the physiological level and position coding for letters in words (e.g. see Pammer et al., 2004). Second, visual attention may impact the visible window of text while reading. Ans, Carbonnel and Valdois (1998) developed a connectionist model of reading that uses a visual attention window, wherein maximal parallel processing occurs, and which is modified in size based on reading mode. The attentional window in this model can be narrowed to focus on smaller portions of orthographic input if needed. This notion of a reader's attentional control over the number of letters processed at a glance may be related to the concept of visual span, the number of letters one can recognise at a glance (O'Regan, Levy-Schoen & Jacobs, 1983). Legge et al. (1997, 2001, 2002) systematically studied visual span size with regard to print size, and found that visual span decreases with very small and very large letters, and that reading speed covaries with the size of the visual span. Normally, smaller visual spans (in characters) are found for print sizes smaller than the critical print size. The present finding of a difference in curve shape for dyslexic readers could be because of a difference in their visual span. That is, we would predict that dyslexic visual spans may continue to increase with increasing print size to a somewhat larger print size (corresponding to the increased critical print size), and the size of the visual span may be smaller than normal even above the critical print size. These potential differences in visual span may be related to inter-letter effects such as crowding or attentional resolution.

While each of these theoretical accounts – position coding, crowding, visual attention or visual span – could contribute to the present print size effect for dyslexic readers, the theories are not mutually exclusive; rather they are likely to be related. For example, imprecise letter position coding could result from attentional disturbances or increased crowding effects in dyslexia. Huckauf and Heller (2002) demonstrated that crowding effects could be explained by position uncertainty rather than interactions of adjacent letter forms. Teasing apart letter position coding and crowding effects on dyslexic reading would require further study by independently manipulating print size and letter spacing. Furthermore, both position uncertainty and crowding effects may contribute to the extent of one's visual span for reading.

It is also possible that a magnocellular deficit could be manifest as slower reading with small text in the present study because of oculomotor factors. The magnocellular visual pathway has neural connections with areas involved in eye-movement programming (e.g. the superior colliculus). Because reading, as in the present study, requires a succession of saccadic eye movements, a magnocellular deficit could have an impact at this level instead of or in addition to a letter-position coding deficit. Producing eye movements within text with small print may be especially difficult, as Kowler and Anton (1987) found that very small saccades may take longer to produce. The present study cannot speak to this possibility. An investigation comparing print size effects of text presentations with and without required eye movements (e.g. normal page-formatted text versus rapid serial visual presentation (RSVP)) would be required to differentiate these effects.

Because the dyslexic group was older in the present sample, one could speculate that the characteristic differences of the two-limb curves result from a general developmental lag, where dyslexic readers behave like younger non-dyslexic readers with regard to having slower maximum reading rates and higher critical print sizes. That the groups still differed when reading level was controlled runs counter to this claim. We chose to ‘equate’ the reading groups on isolated word identification, although other factors may be developmentally relevant as well. Furthermore, a developmental lag model would be strongly supported if the dyslexic and non-dyslexic data fell on the same critical-print size-by-maximum-reading-rate regression line, but this was not the case either. The dyslexic group showed a lack of a correlation between these variables, whereas the nondyslexic group showed a fairly strong correlation. This implies some de-coupling of the maturational influences that normally tie the two variables together. Here, the slowest dyslexic readers did not necessarily require larger print size thresholds to attain their maximum reading speed, whereas slower non-dyslexic readers did.

One might also expect that if general intelligence accounts for sensory processing differences in dyslexia, then those with lower IQ would show the greatest impact on sensory processing (i.e. they would have higher critical print sizes). However, this did not appear to be the case with the group we tested. It was the children with higher IQ and an IQ-achievement discrepancy who actually had the highest critical print sizes.