In acoustic echo cancellation (AEC), the echo paths are often long, sparse and rapidly changing. In this paper, a new gradient controlled proportionate adaptation technique is proposed for AEC. A time-averaging gradient estimate is assigned as the gain distribution vector to assign individual step sizes so that the filter taps are updated proportionally to the magnitude of the mean of the gradient vector. Based on this new proportionate technique, we propose the gradient-controlled improved proportionate affine projection algorithm (GC-IPAPA), and gradient-controlled improved proportionate normalized least mean square algorithm (GC-IPNLMS). Simulation results demonstrate significant improvements in convergence rate and a more robust performance compared with traditional proportionate algorithms for AEC.