The purpose of this study was to investigate the collaborative dialogue patterns of pair programming and their impact on programming self-efficacy and coding performance for both slow- and fast-paced students. Forty-six postgraduate students participated in the study. The students were asked to solve programming problems in pairs; those pairs' conversations were recorded when they worked on their tasks. Data analysis methods, including lag sequential analysis, cluster analysis and paired t-test, were employed, and the results showed that (1) four collaborative dialogue patterns emerged: Lecture, Guide, Question and Answer (Q&A), and Inquiry patterns; (2) Guide and Inquiry patterns significantly increased programming self-efficacy for both fast- and slow-paced students while Lecture and Q&A patterns significantly increased programming self-efficacy for slow-paced students but not for fast-paced students; (3) Guide and Inquiry patterns played a significant role in improving coding performance for slow-paced students. The study reveals a complex relationship between collaborative dialogue patterns with programming self-efficacy and coding performance, critically affecting students' pair programming quality. Further details of the findings are also discussed.