This paper aimed at providing a simple and direct method for shortening the lengthy 108-item Fennema-Sherman Mathematics Attitude Scales (FSMAS). The method proposed uses the trace information (widely used in other popular procedures such as principal component analysis and regression analysis) as the criterion for item selection. Results for half-length FSMAS versions (54 items) constructed using the trace-information criterion compared favourably with Mulhern and Rae's (1998) 51-item FSMAS short form. But unlike the later hort form, our method retained as much variance as possible in the original 108-items and maximized the correlations, and hence predictive validity, with the full-length FSMAS version. The appropriateness of selecting items from a larger item pool based on employing a trace-information criterion is discussed within the context of the domain-sampling model.

Key words: BI-method, domain sampling, item selection, stepwise, trace-information, item selection