how to improve data fit accuracy with neural network?

I have a one hidden layer network with 6 inputs and 4 outputs. I use neural network tool to perform 'Function Approximation and Nonlinear Regression' training . I use 'trainlm' and 'trainbr' as trainFcn, try different numbers of neurons for hidden layer and change other options which I think can improve training result. But no matter what I do, I cannot get R value over 0.9 for trainingg, validation,test and all, which means the outputs cannot fit targets well. Does my data are not good enougth, therefore results in poor fitting. Who can provide me some suggestion, or what should I do? Thank you.

[Greg Heath]

How many examples?

Maximum number of hidden nodes?