Produce quality software inside expected time and low cost has been one of the main challenges in the software industry today. Therefore, it is fundamental make estimates of size, effort, resources, cost and time spent in the software development process. Predictive models such as models based on analogy can be an alternative, especially in small and medium sized software development, they need perform reliable estimates. In this paper, we propose a methodology for pre-processing of data for use in software effort estimation. The results of use this methodology, applying Case Based Reasoning - CBR, indicate that the pre-processing enables to improve the accuracy of estimates.