The 32P-postlabeling assay is widely used for the analysis of DNA adducts. Some adducts can be detected with very high sensitivity but quantification can be unreliable, particularly if it is based only on comparison with unmodified nucleotides (relative adduct labeling, RAL values). Furthermore, guidelines to calculate detection limits for adduct concentrations are lacking. This is particularly important for human biomonitoring studies of environmental exposures, where a low adduct level can remain undetected. Reports of null results of toxicity studies should always include a limit of detection, indicating the effect magnitude that would have produced, with a given probability of false negative (type II error), a statistically significant increase (type I error). Here, we report on a procedure based on t-statistics to calculate two types of detection limits, the "critical level (CL)" and the "detection level (DL)". The first is the size of the difference between exposed and controls required to achieve statistical significance. The second is the size of the difference that will be detected with a chosen probability of a false negative. For the degrees of freedom (d.f.) to be used for the t-values, a general formula is given so that different standard deviations and group sizes of control and exposed groups can be handled. A sample calculation of the whole procedure is shown, using the null data for the formation of a particular adduct in lung DNA of styrene-treated mice, analyzed by 32P-postlabeling. The procedure takes into account: (i) TLC-specific background radioactivity; (ii) variability within the control and exposed groups; and (iii) confidence limits for the factor to convert 32P-radioactivity to amounts of adduct. The latter step incorporates the variance of the differences between the samples and replicates spiked with adduct standard. A statement such as follows is the result: the concentration of the alpha-isomer adduct of styrene 7,8-oxide at the O(6)-position of guanine in mouse lung DNA would have to be at least 12 adducts per 10(8) nucleotides to be detected in the given experiment on a 5% level (type I error), with a probability of 5% to miss an existing effect (type II error).