Multiscale error diffusion (MED) is superior to conventional error diffusion algorithms as it can eliminate directional hysteresis completely. However, due to its frame-oriented processing nature, the computational complexity is comparatively high. Furthermore, though theoretically MED can remove directional hysteresis by eliminating predefined scanning paths and causal filters, no comprehensive quantitative analysis on this issue can be found in the literatures. In this paper, a fast MED algorithm is proposed and a detailed analysis on the performance of various MED algorithms including the proposed one are provided. Analysis and simulation results show that the proposed algorithm can reduce the computational complexity without sacrificing the image quality as compared with conventional MED algorithms. The proposed algorithm also supports parallel processing and hence can further reduce the processing time.