Physiological rhythms arise from nonlinear interactions between biological mechanisms and environmental conditions. A possible approach to study these dynamics is by means of simplified mathematical models. An essential aspect of these models is how to determine the statistical significance of the rhythms present in a temporal series. The aim of this work is to propose an automatic rhythm analysis method based on lasso or l 1-regularized linear regression, with physiological rhythm components as features. These models have sparse solutions, allowing to identify relevant rhythms. Since the sine and cosine components of a given period constitute a natural group structure, we used a group lasso model. A cross-validation scheme preserving the temporal structure of the signal allowed to select the regularization parameter. Synthetic signals were used to test the method, combining different sinusoidal rhythm components plus gaussian noise. The method was also applied to study the rhythms in heart rate signals (HR). The method correctly detected 98% of rhythm patterns on the synthetic data. The method was also able to extract significant cardiac rhythms in HR signals. Since lasso is the closest convex relaxation of the best feature subset selection problem, the proposed method is able to optimally identify the rhythms present physiological signals.