Gait abnormalities can be studied by means of instrumented gait analysis. Foot-switches are useful to study the foot-floor contact and for timing the gait phases in many gait disorders, provided that a reliable foot-switch signal may be collected. Considering long walks allows reducing the intra-subject variability, but requires automatic and user-independent methods to analyze a large number of gait cycles. The aim of this work is to describe and validate an algorithm for the segmentation of the foot-switch signal and the classification of the gait cycles. The performance of the algorithm was assessed comparing its results against the manual segmentation and classification performed by a gait analysis expert on the same signal. The performance was found to be equal to 100% for healthy subjects and over 98% for pathological subjects. The algorithm allows determining the atypical cycles (cycles that do not match the standard sequence of gait phases) for many different kinds of pathological gait, since it is not based on pathology-specific templates.