Nowadays there are many models for software development cost estimation, providing project managers with helpful information to make the right decisions. One of such wellknown mathematical models is the COCOMO model. To estimate costs and time, this model uses coefficients, which were determined in 1981 by means of the regression analysis of statistical data based on 63 different types of project data. Using these coefficients for a modern project, the appraisal may not be accurate; therefore, the aim of this paper is to optimize the model coefficients with genetic algorithms. Genetic algorithms are evolutionary methods for optimization. To evaluate population, the genetic algorithm will use a set of descriptive attributes of several software development projects. These attributes are the number of lines of a code, costs and implementation time of a project. Project costs estimated by means of the COCOMO model will be compared with the real ones, this way evaluating the fitness of an individual in the population of possible solutions.