Reservoir engineering calculations required reservoir fluid viscosity data. In the absence of reliable experimental data, empirically derived correlations are used to predict the PVT property. However, several viscosity correlations in the literature were developed using regional data with limited accuracy and applicability. Artificial neural network (ANN) is a computer model that attempts to mimic a simple biological learning process and simulate specific functions of human nervous system, a new computation algorithm that could be used to develop more robust and reliable model for generating PVT data. It is an interconnection of nodes, called neurons. The authors have evaluated the industry widely used viscosity correlations using Nigeria crude oil data and explored the use of neural network in estimating viscosity at bubblepoint pressure and dead oil viscosity with the aim of getting a more accurate oil viscosity predictive model compared to widely used correlations in the literature. This study used the ANN backward propagation procedure with the Levenberg-Marquardt algorithm for the optimization procedure for both the viscosity at the bubblepoint pressure and dead oil viscosity models. The number of data set used for viscosity at the bubblepoint pressure is 1809 and dead oil viscosity is 1750. A number of neural network hidden layer designs were considered and tested. Each successful trained model was tested to ensure that overfitting does not occur and can predict output from the inputs that were not seen by the model during training. Sixty percent of the data was used to train the network, twenty percent to cross-validate the relationships established during the training process and the remaining twenty percent to test the model. The results of artificial neural network model for viscosity at bubblepoint show that the model gives higher accuracy compared to the published correlations with average absolute relative error of 11.05 and coefficient of correlation of 0.98. Besides, the dead oil viscosity neural network model shows a substantial improvement over correlations with average absolute relative error of 12.6 and coefficient of correlation of 0.91. Copyright 2017, Society of Petroleum Engineers.