Confidence (i.e., efficacy) plays a pivotal role in sport, whether it is athletes' confidence in themselves, their coaches, or in their collective group. Studies have shown that athletes' self-efficacy, other-efficacy, and collective efficacy beliefs lead to beneficial outcomes, such as enhanced performance, effort, and persistence. While athletes' perceptions of coach behaviors are associated with their efficacy beliefs, no known research has utilized observed coaching behaviors when exploring this relationship. The purpose of this study was to directly observe coaches and examine how their behaviors (i.e., instruction and feedback, positive and negative evaluation, autonomy support, and motivational climate) predict athletes' self-efficacy, other-efficacy, collective efficacy, and relation-inferred self-efficacy (RISE) beliefs. The sample consisted of three coaches, and 64 athletes from NCAA Division I and III sports. Teams included Division I Women's Soccer (n = 19), Division III Men's Lacrosse (n = 35), and Division III Football (n = 10). Coaches were recorded at a single practice, and their behaviors were coded using the Assessment of Coaching Tone observational coding system (Erickson and Cote, 2015). Athletes completed surveys measuring their self-efficacy, other-efficacy, collective efficacy, and RISE. Initial analyses showed that the coaches differed in behavior use. The football coach directed the most behaviors toward individuals, the soccer coach was the most autonomy supportive, and the lacrosse coach was the most mastery oriented. Furthermore, the football team reported higher self-efficacy than the soccer team, and higher collective efficacy than both the soccer and lacrosse teams. Multiple regression analyses revealed that none of the hypothesized coach behaviors were unique predictors of any forms of athlete efficacy beliefs. The findings contribute to existing literature on efficacy sources and provide directions for future research.