Recognizing gait of people has been of great interest to the researchers of biometrics in the last decade. The robust features have been recently developed to identify human’s gait under different conditions. But developing efficient gait template preserving spatio-temporal features of walking is still an open problem. To address this issue, we develop a patch-based feature that can describe rhythm of walking under covariate factors properly. In our method, a new gait signature (i.e. set of spatio-temporal features) is computed from distribution of local patches in a sequence. The given signature has been used to adjust the weights of spatio-temporal coordinates and the corresponding weights are concatenated with the Gabor features. As a result, a new augmented template called Patch Gait Feature (PGF) is derived accordingly. In addition, to verify how our feature template is efficient in gait recognition, we apply two common classification methods (PCA + LDA and Random Subspace Method (RSM)) separately and evaluate the results under different challenging conditions. The recognition rate on the USF dataset indicates Rank1/Rank5 accuracies of 61.59/80.67% with PCA + LDA and 76.01/86.59% with RSM and shows an improvement of about 5% with rational computational complexity compared with other related methods.