In this paper, a data mining technique for protein sequence pattern extraction is developed. Specifically, the aim is to explore the use of association rules as a basis to build successful secondary structure predictors, in a sequencestructure layer. No heuristic or biological infor mation is taken into account in the present study and only the information given by the association rules is used as a basis for building a secondary structure predictor. This work gives some insights about secondary structure prediction features to be used in learning algorithmsI¾ this is expected to be useful to achieve substantial improvements of accuracy in protein secondary structure prediction.