This thesis presents a study of on-site labor productivity in building construction using the work sampling method. The study is based on a field investigation of a number of selected construction operations on three buildings in Montreal, Quebec, Canada. The developed models revealed related parameters' impact on labor productivity. Neural network was used as a method for the development of the models presented in this thesis. The developed models are based on the data collected using work sampling and were developed using NeuralShell2 software. The network was trained and tested using 221 data points collected from real construction projects that were performed in Montreal in a 30-month period. The models' development and validation utilize real-world data from the projects. Three types of neural network-based models were developed. The first type of models is back propagation neural network (BPNN) models associated with different settings. The fifth model has shown the best results.