A simulation procedure applied to select a suitable statistical procedure to model flood flows in Rufiji River basin in Tanzania is presented. In this paper, a flood frequency procedure refers to the combination of a probability distribution with a suitable method of parameter estimation. A predictive ability test based on bias between observed and predicted flows is used to select the suitable procedure. Based on the results of the simulation, the most suitable statistical frequency procedures for modelling flood flows for the homogeneous regions of Rufiji basin seem to be log-Pearson type 3/method of moments (LP3/MOM) and Pearson type 3/method of probability weighted moments (P3/PWM). The two procedures gave the lowest bias values for different return periods and sample sizes considered in the sUidy. The study is based on the data from 42 sites in the basin