When planning an affected sib pair collection for use in a genomewide search for complex trait loci, researchers must ask: (a) Which family structures will yield the most informative pairs? and (b) Should recruitment extend beyond the index sib pair? The optimal collection strategy will depend on the trait's genetic architecture, but this is rarely known for non-Mendelian diseases. In the present report, we study the consequences of collecting only those sib pairs arising from pedigrees with a precisely specified structure as opposed to a strategy that collects all affected sib pairs at random (i.e., blind to the affection status of first-degree relatives). The former approach turns out to be risky because the power of specific pedigree structures can vary dramatically even among models producing identical observable parameters (such as population prevalence and sibling recurrence rate). In contrast, the latter approach typically involves only a modest loss of power as compared with the optimal (but unknowable) design. Further, we compare the strategy of collecting all affected sib pairs at random with the alternative of imposing some modest limitations on family structure (e.g., presence of at least one unaffected sib or parent). The latter approach generally provides some increase in power but entails additional effort to contact and phenotype relatives: the overall merit of imposing such requirements needs to be evaluated in the context of the specific disease to be studied and of the clinical and analytical resources available. In addition, these findings suggest that a further explanation for failure to replicate positive complex trait linkages lies in differences in ascertainment strategy between data sets.