"At the same time, however, symptom-focused treatment strategies may fail to address the personality structure that provides a context for understanding disordered eating. Patients whose personality profiles match the overcontrolled, constricted prototype, for example, rarely recognize their stance toward their own impulses and relationships as a problem. What brings them into treatment is typically someone else’s concern about their weight. If their attitudes toward their needs and feelings in general (and not just toward food) do not become the object of therapeutic attention, they are likely to change with treatment from being starving, unhappy, isolated, and emotionally constricted people to being relatively well fed, unhappy, isolated, and emotionally constricted people. The data also raise questions about the extent to which axis II is adequate for describing clinically meaningful patterns of personality pathology, at least for women with eating disorders. Patients in the high-functioning/perfectionistic cluster generally lacked diagnosable axis II pathology; indeed, in our study (as in the other studies that have isolated a similar cluster), they were defined by the absence of such pathology. These patients are articulate, conscientious, and empathic, and they tend to elicit liking in others. Yet they clearly have personality pathology—that is, enduring, problematic patterns of thought, feeling, motivation, and behavior. They are self-critical, perfectionistic, competitive, anxious, and guilt-ridden, and these aspects of their personality require clinical attention. The data reported here make sense in light of other findings that roughly 60% of patients treated for clinically significant personality pathology do not have problems severe enough to be diagnosable on axis II and that their personality problems (e.g., perfectionism and chronic feelings of guilt) generally are not reducible to any axis I syndrome (21, 22). Available data suggest that these patients represent the majority of patients treated in clinical practice and are not simply the “worried well.” Either axis II needs to be expanded from a personality disorderaxis to a personalityaxis that includes the range of functioning (from relatively healthy to relatively impaired), or subtypes such as those uncovered here need to be built into axis I. From a methodological standpoint, the results of this study suggest that we should routinely test for subtypes in our data sets rather than assuming homogeneity of categories. Group means may not be very meaningful when substantial intracategory heterogeneity exists, particularly if this heterogeneity is ordered, not random. The problem is particularly pronounced if pathology can be expressed in phenotypically opposite directions, leading to means that cancel out patterned within-group variability. Thus, although the etiological data on sexual abuse reported here are correlational and preliminary, they suggest that the same risk factor—sexual abuse—may manifest in opposite personality and behavioral styles—constriction and inhibition on the one hand, and dyscontrol and promiscuity on the other. Whether this is true of other etiologically significant psychosocial variables, such as harsh parental criticism (which, from a clinical point of view, appears sometimes to lead to self-criticism, sometimes to hostility and criticism toward others, and sometimes to both in adulthood), is an important question for future research."