Measurement of thresholds for warm and cold sensation was performed on 106 normal subjects, at thenar eminence and foot dorsum. Three test algorithms were used, the reaction-time-inclusive method of limits, and reaction-time-exclusive methods of levels and staircase. Tests were repeated 2 weeks following the first for most of the subjects, and after elimination of 5 outlying subject data points, and determination of no systematic relationship between magnitude and variability of test scores, data from between 72 and 76 subjects were used to derive repeatability coefficients, by ANOVA-based procedures which extend standard repeatability assessment methods. Normative data tables are presented, with measures of repeatability for the various algorithms and modalities. Method of limits tests exhibited inter-session bias, and large repeatability coefficients, compared with methods of levels and staircase, which exhibited no bias and had better (lower) repeatability coefficients. All three methods had similar test durations. We conclude that on the basis of these data, the reaction-time-exclusive methods of levels and staircase have a definite advantage over the method of limits.