This paper describes a method of clothing-invariant gait recognition by modifying intensity response function of a silhouettebased gait feature. While a silhouette-based representation such as gait energy image (GEI) has been popular in gait recognition community due to its simple yet effective property, it is also well known that such a representation is susceptible to clothes variations since it significantly changes silhouettes (e.g., down jacket, long skirt). We therefore propose a gait energy response function (GERF) which transforms an original gait energy into another one in a nonlinear way, which increases discrimination capability under clothes variation. More specifically, the GERF is represented as a vector of components of a lookup table from an original gait energy to another one and its optimization process is formulated as a generalized eigenvalue problem considering discrimination capability as well as regularization on the GERF. In addition, we apply Gabor filters to the GEI transformed by the GERF and further apply a spatial metric learning method for better performance. In experiments, the OU-ISIR Treadmill dataset B with the largest clothing variation was used to measure the performance both in verification and identification scenarios. The experimental results show that the proposed method achieved state-of-the-art performance in verification scenarios and competitive performance in identification scenarios.