OBJECTIVE Short-wavelength automated perimetry (SWAP) has gained popularity as a clinical tool for the assessment of short-wavelength-sensitive (S)-cone visual function, but has also been shown to have higher threshold variability than conventional achromatic perimetry, possibly due to an imbalance between S-cone adaptation and long (L)- and medium (M)-wavelength-sensitive cone adaptation. To investigate potential causes for this relatively high variability, we studied the effects of luminance and S-cone adaptation on variability in S-cone increment thresholds. METHODS Foveal S-cone increment thresholds were measured on adapting backgrounds ranging from 1.17 to 4.17 log troland (Td) and from -0.16 to 3.66 log S-cone trolands (Td(S)). Within-session variability (slope of the psychometric function) was evaluated in 2 trained and 15 inexperienced normal observers. Test-retest variability was evaluated in the 2 trained observers, and interobserver variability in the group of 15 observers. Multiple linear regression was used to model the effects of log luminance, log S-cone adaptation, and second-site polarization (ratio of luminance and S-cone adaptation; log [Td/Td(S)]). RESULTS Test-retest variability was lower for conditions with higher levels of S-cone adaptation (F = 9.04, P = 0.013, for the trained observers). Adaptation conditions with lower levels of polarization were associated with lower within-session variability (F = 6.9, P = 0.011; trained observers) and interobserver variability (F = 33.7, P = 0.004; group of 15 observers). CONCLUSIONS Variability of S-cone increment thresholds can be reduced by using adaptation conditions with a higher level of S-cone adaptation and/or a more balanced ratio between luminance and S-cone adaptation than is used for SWAP.