The SA filters encompass a large class of filters based on order statistics as well as linear FIR filters. The class of SA filters is defined as a multi-stage filters whose output is linear combination of nonlinear(minimum, maximum, exclusive-OR) sub-filter outputs. According to the lust stage nonlinear sub-filters, SA filters are called SAMAX, SAMIN, and SAXOR filters. In this paper, optimal SAMAX and SAMED filters are designed for images corrupted by biased noise. The performance analysis of this experiment shows that SAMAX filters outperforms SAMED filters for biased noise. In the case of un-biased noise, the SAMAX and SAMED filters give the same performance. This result leads us to a new guideline in the application of SA filters.