Abstract. We present a new method for computing the scalar multiplication on Koblitz curves. Our method is as fast as the fastest known technique but requires much less memory. We propose two settings for our method. In the first setting, well-suited for hardware implementations, memory requirements are reduced by 85%. In the second setting, well-suited for software implementations, our technique reduces the memory consumption by 70%. Thus, with much smaller memory usage, the proposed method yields the same efficiency as the fastest scalar multiplication schemes on Koblitz curves.