Discovering patterns from sequence data has significant impact in genomics, proteomics and business. A problem commonly encountered is that the patterns discovered often contain many redundancies resulted from fake significant patterns induced by their strong statistically significant subpatterns. The concept of statistically induced patterns is proposed to capture these redundancies. An algorithm is then developed to efficiently discover non-induced significant patterns from a large sequence dataset. For performance evaluation, two experiments were conducted to demonstrate a) the seriousness of the problem using synthetic data and b) top non-induced significant patterns discovered from Saccharomyces cerevisiae (Yeast) do correspond to the transcription factor binding sites found by the biologists. The experiments confirm the effectiveness of our method in generating a relatively small set of patterns revealing interesting, unknown information inherent in the sequences.