A Neural network (NN) to predict serologic specificity of HLA alleles was first introduced in 2003 by Maiers et al. The procedure involved several manual steps and prediction errors were estimated from an arbitrarily selected single partition of available labeled data. However, these might not reflect the true predictive capability of the model. Our aim was to provide an automated framework for the existing NN, estimate the true prediction accuracy and capability using a cross-validation model and compare with other predictive machine learning techniques.An NN framework was created and validated in the R software suite to predict serologic specificities for HLA-A, B, C, DQB1, DRB1 and DPB1. We established a method to tune NN parameters using a grid technique with leave one out cross validation and compared the prediction accuracy of the NN with a k-Nearest Neighbors (kNN) technique.A concordance between the manual and automated NN of upwards of 96% was obtained for all loci, measured on an independent unlabelled test dataset. With parameter tuning techniques for the NN we have estimated the average prediction accuracy across all loci to be 86% .We also observed that kNN provided better prediction accuracy of 90% (average). These numbers were slightly lower than the previous NN validation based on a single partition of the data, but still provide acceptable performance for both the NN and kNN.Human involvement in NN model selection is both extremely time consuming and prone to errors. An automated framework minimizes the risks of over-training by monitoring variations in errors. Cross-validation techniques are known to provide better generalization error estimates. In this context, our reported error estimates maybe closer to the actual error rates one can achieve with similar models. We also demonstrate kNN outperforms the existing NN. This result should support efforts in finding relationships between alleles and associated serologic specificities.