neural-network- how to choose the best model

I am using nftool to train my data (90 piece of data), and my training algorithm is "Bayesian Regularization" I am going to use different neurons to compare and find which one is the best model. Is there any appropriate step I need to follow? Do I need to retrain again and again until getting the good performance of model for each neurons?