This study investigated the extent to which self-efficacy and motivation served as a predictor for mathematics achievement of fifth grade students in United Arab Emirates (UAE) across gender and achievement levels. Self-efficacy was measured by two scales, which differed in levels of specificity--Category Specific and Task Specific. Motivation was measured through four sub-constructs of motivation-- Amotivation, External Regulation, Introjected Regulation, and Intrinsic Motivation. A total of 287 fifth grade students with an average age of 10.3 years were randomly selected to participate in this study. The multiple regression model showed that the six predictors were able to explain together high percentage (32%) of the variance of mathematics achievement. Also the results indicated that the best three predictors were Task Specific, External Regulation, and Intrinsic Regulation. When conducting the regression model across gender, the results showed that 30% of the variance in mathematics achievement was explained by the six predictors for the male group while only 21% of the variance was explained for the female group. The regression model was not invariant across achievement levels. While the model predicted approximately 20% of the variance of mathematics achievement for each of the low and high achieving students, the model was not statistically appropriate for the medium achievement students as it predicted only 5% of the variance of mathematics achievement. Additionally, the performance of the six predictors varied according to the achievement level.