Many project managers try to predict the defects in software systems before they deliver it to the customer. The purpose is to develop the defect prediction model that can provide better insight to tackle the software quality issues in an efficient way. Method: This study is focused on aneural net model that receives the input as function points based on the size of software and various forms of software. The output is a total number of defect potentials per function point for different sizes of software and different domains. Results: Our neural network based experimental results have prediction error rate as 0.001%, which indicates clear superiority over existing defect prediction models. Conclusion: Neural network with six neurons in the hidden layer achieves faster and better accuracy in prediction. Network with less or more number of neurons in the hidden layer results in more error or iterations. Keywords: Defect potentials, Defect prediction, neural network, Software quality. © 2016, International Journal of Pharmacy and Technology. All rights reserved.