In recent years, cross-disciplinary scientific collaboration has been proved to be promising for both research practice and innovation. Lots of efforts have been spent in collaboration recommendation. However, the cross-disciplinary information is hidden in tons of publications, and the relationships between different fields are complicated, which make it challengeable recommending cross-disciplinary collaboration for a specific researcher. In this paper, a novel cross-disciplinary collaboration recommendation method (CDCR) that unearths the common cross-disciplinary collaboration patterns and historical scientific field preferences of authors is proposed to recommend potential cross-disciplinary research collaboration. In CDCR, a research field discovery algorithm is designed to classify scientific topics obtained from the publications into the correct field automatically. Then, the collaborative patterns are studied through analyzing the composition fields and the corresponding percentage of all publications. Furthermore, we investigate the common correlation of different research fields. Based on the common correlation and the researcher's specific pattern, the most valuable fields will be listed by CDCR. The effectiveness of our approach is evaluated based on a real academic dataset.