Software development effort estimation is fundamental part of software project management. It is the process used to predict the most probable effort required to perform specific work. Based on forecasted effort it is possible to determine costs and allocate required resources. The effort estimation inherently includes various factors and therefore the process of decision making and producing the predictions regarding required efforts is in its nature a process of reasoning with uncertainty. To enhance this process software engineers are using various approaches, application of data mining and knowledge discovery techniques proved to be especially effective. This paper reports a study in which Bayesian networks (BN) are used to improve software development effort estimation. Study examines tree major entities involved in estimation process – projects, work items and estimators. The analysis is based on real data collected from software projects executed in Croatian software company. Study found that Bayesian networks are especially suitable for modeling of effort estimation and can significantly contribute to management of software projects.