Using data from the Longitudinal Study of American Youth (LSAY), we aimed to determine the causal ordering between mathematics anxiety and mathematics achievement. Results of structural equation modelling showed that, across the entire junior and senior high school, prior low mathematics achievement significantly related to later high mathematics anxiety, but prior high mathematics anxiety hardly related to later low mathematics achievement. Mathematics achievement was more reliably stable from year to year than mathematics anxiety. There were statistically significant gender differences in the causal ordering between mathematics anxiety and mathematics achievement. Prior low mathematics achievement significantly related to later high mathematics anxiety for boys across the entire junior and senior high school but for girls at critical transition points only. Mathematics anxiety was more reliably stable from year to year among girls than among boys.