Abstract We investigate the use of the Probabilistic Incremental Programming Evolution (PIPE) algorithm as a tool to construct continuous cumulative distribution functions to model given data sets. The PIPE algorithm can generate several candidate functions to fit the empirical distribution of data. These candidates are generated by following a set of probability rules. The set of rules is then evolved over a number of iterations to generate better candidates regarding some optimality criteria. This approach rivals that of generated distribution, obtained by adding parameters to existing probability distributions. There are two main advantages for this method. The first is that it is possible to explicitly control the complexity of the candidate functions, by specifying which mathematical functions and operators can be used and how lengthy the mathematical expression of the candidate can be. The second advantage is that this approach deals with model selection and estimation at the same time. The overall performance in both simulated and real data was very satisfying. For the real data applications, the PIPE algorithm obtained better likelihoods for the data when compared to existing models, but with remarkably simpler mathematical expressions.