Schedule, effort and cost are three important elements of a successful project management. When it comes to predict them; making decisions based on only daily situation assessment reports is not enough anymore. Decision-making becomes more difficult in the case of uncertainty. However, administrators gain competitive advantage with the help of machine learning. in this study, it is aimed to predict software project effort using Artificial Neural Networks (ANNs). Unlike the other studies in the literature, authors preferred to evaluate the performance of ANNs from the perspective of one-class and binary classification, rather than estimate the exact project effort (person in months). For this purpose, the target attribute "effort" is categorized into two distinct groups using equal frequency discretization method. Analysis are performed on a popular dataset named COCOMA 81 which is used by many researchers for effort estimation. Analysis are performed with R Programming Language and RStudio. Model performances are evaluated using 5-fold cross validation technique. In this study Multi-Layer Feed-Forward (MLFF) ANNs with Back Propagation Algorithm and Autoassociative Neural Networks (AANNs) is used as binary and one-class classification algorithms respectively. The optimum ANN architecture with the highest accuracy (0,72) and the lowest error (0,28) is obtained with MLFF ANNs.