We present an iterative algorithm that uses randomness and statistical techniques to improve existing met,hods for recognizing protein structural motifs. Our algorithm is particularly effective in situations where large num-bers of sufficiently diverse examples of the motif are not known. These are precisely the situations that pose sig-nificant difficulties for previously known methods. We have implemented our algorithm and we demon-strate its performance on the coiled coil motif. We test our program LearnCoil on the domain of 34randed coiled coils and subclasses of 2-stranded coiled coils. We show empirically that for these motifs, our method over-comes the problem of limited data. 1