Astigmatic errors (0.50 D to 1.50 D in 0.25 D steps) were induced in each of 5 subjects, and cylinder axis thresholds were assessed from 12 reversals of a 2.5 degrees stepsize staircase for 0.50 D and 1.00 D cross cylinders (CC). Equivalent power thresholds were 0.077 D (0.05 SD) for the 0.50 D CC and 0.104 D (0.05 SD) for the 1.00 D CC. Findings are significantly different from predictions of a "linear" blur detection model, but are consistent with a model assuming constant blur thresholds and power-dependent distortion thresholds. Our results support the use of minimal CC power in refining cylinder axis.