Most functional RNA molecules have characteristic, highly conserved structures, such as pseudoknots. But the prediction of RNA pseudoknots has largely remained a difficult problem, and many existing algorithms for prediction of RNA secondary structures do not have the ability to predict pseudoknots. Here we present a new method for predicting RNA secondary structures including pseudoknots through iteration. The algorithm combines thermodynamic and covariation information to assign scores to all possible base pairings. Base pairings are then predicted with the help of the iterated RNA folding algorithm based on minimum of free energy. Test result shows that nearly all pseudoknots are predicted. Compared to other methods, the method achieves a specificity that is among the best and a sensitivity that is nearly the best.