Factor analysis can be a valuable tool for assessing validity; however, the reliabilities of identified factors must be attended to. This study examined the impacts of item variance on factor stability and structure. The attempt was to determine empirically whether unnumbered graphic scales scored using a larger number of scale steps produce more stable factors than do unnumbered graphic scales scored using a smaller number of scale steps. Inservice teachers read and rated their preference of summaries of 16 models of teaching. The scales were rescored using five different numbers of scale steps. Results suggest that researchers need not be concerned that an overly-specific scoring metric will distort the meaning of the identified factors. In addition, it is concluded that unnumbered graphic scales are a promising method of collecting attitudinal data, particularly if the scales are scored with a reasonably large number of scale steps. (Author/GK)