Project size, as measured by the amount of investment required, is a relevant parameter to be used in project selection. The evaluation of a project portfolio must consider the variety of project sizes that may be met, so that a proper model should be adopted to describe that variety, especially for its use in simulation. In this paper, a log-normal probability model is suggested to describe the dispersion of project sizes within a project portfolio. The model is obtained on the basis of two real datasets spanning over ten years of observations, and after comparison with competing Gamma and Pareto models. The parameters of the log-normal model are provided as resulting from the best-fit procedure, and indications are also given for the values to use in a simulation study.