The Mathematics Attitude Scale (Aiken, 1974) purports to reflect attitude toward mathematics on two basic dimensions, enjoyment and value. While this scale is widely used in mathematics research, the only empirical evidence supporting the two-factor structure is the item analysis of college freshmen responses. The present study examines the suggested factor structure for 96 elementary teachers. The responses of the teachers were subjected to both item analysis procedures and selected factor analysis algorithms. Because of the exploratory nature of this research, principal component and classical factor analyses (using squared multiple correlation coefficients as initial estimates of communality) were employed in conjunction with several rotation procedures. The resulting three-factor solution (enjoyment, general value, and personal value) is more meaningful both psychologically and empirically than the original two-dimension scale, given a sample of in-service teachers. While internal consistency estimates for the first two factors were only slightly improved by imposing the new structure, a meaningful third factor with an internal consistency of .78 emanated. Moreover, increases in item discriminination indices were found. (Author/RC)