Anaerobic threshold (AT) during exercise is usually noninvasively determined by assuming a two-segment mathematical relationship between two ventilatory parameters. In the literature, all the possible pairs of segments are first considered, and the most appropriate pair is then selected according to at least-squares method. In such a model, the AT is considered to be related to the joining point of the two segments. In order to test the reliability of the model, we compare the results of the least-squares method to those based on maximum probability method in discriminating the two regression coefficients. In order to test the reproducibility of the two different criteria, comparisons have been repeated after data have been filtered. A paired t test was used to carry out comparisons. Ventilatory parameters were collected in 10 healthy subjects during the use of a bicycle ergometer. The required power was increased every 15 s by steps of 30 W, starting from 50 W. Ve, VO2 and VCO2 have been sampled every 15 s, then the three functions--Ve versus VO2, Ve versus VCO2 and VCO2 versus VO2--were considered. Each function was stylized with two linear segments. Each segment was estimated by using a second-kind linear fitting. We verified that: (i) the AT may be reliably appreciated depending on the pair of selected parameters; (ii) only when data are smoothed is no difference between the two criteria documented (Ve vs. VO2, p = 0.99; Ve vs. VCO2, p = 0.54); (iii) no significant difference, related to smoothing, is documented both in using the least-squares method (Ve vs. VO2, p = 0.61; Ve vs. VCO2, p = 0.15) and the maximum p level criterion (Ve vs. VO2, p = 0.59; Ve vs. VCO2, p = 0.19).