Abstract: Speedup and efficiency are widely used to evaluate the performance of parallel computing systems. However, they work well only in homogeneous parallel systems. The introduction of heterogeneity in workstation clusters makes them not suitable for parallel computing. To cope with this problem, the heterogeneity of workstation clusters is quantified,and the speedup and efficiency are redefined on the basis of two matrix multiplication experiments. These new models are general enough to cover performance evaluation of both homogeneous and heterogeneous computations.