Best training and bad prediction !!!

I used MLP(fitting) artificial neural network(ANN) to predict solar irradiation, my problem is i have gotten best trainig and testing (R(Training)=0,997 ,R(Test)=0,995, MSE=0,057),but if i want to predict the new values unseen by ANN i got the bad results!! ?? please help me how can i fix this problem??

best regards

[Greg Heath]

Insufficient info

 size(input)?,
 size(target)? 
 Type of net? 
 Number of  delays? 
 Number of hidden nodes?

Greg