The primary purpose of the current study was to use structural equation modeling to examine the relations among background variables (socioeconomic status, prior mathematics achievement), motivation variables (self-efficacy, task interest, school connectedness), self-regulated learning (SRL) behaviors, and performance in middle school mathematics courses. Of particular interest was examining the mediation roles of both self-efficacy and SRL behaviors. Data about three types of motivation beliefs (self-efficacy, task interest, connectedness) were obtained from 331 middle school students using self-report questionnaires, while information regarding student SRL behaviors was obtained from teacher ratings. Structural equation modeling analyses revealed an acceptable fit of the data to the proposed model. In addition to the overall model explaining 51% of the variance in mathematics performance, a key finding was that both cognitive (i.e., self-efficacy) and behavioral (i.e., SRL) latent factors served as key mediators in the model, with each of these factors exhibiting unique effects on mathematics performance after controlling for prior achievement. Furthermore, each of the three motivation beliefs played an important role in the model, particularly regarding the explanation of SRL behaviors. Directions for future research and implications for practice are discussed.