Sib pair tests for linkage to a quantitative trait or affection status are examined by simulation. A sibship of size s contributes only s – 1 independent sib pairs but all s(s – 1)/2 pairs are needed for an efficient test. Redundancy increases with sibship size. Using simulated data on the null hypothesis of no linkage, we have considered equal and unequal weights. Unequal weights lose power as measured by the equivalent number of observations, and goodness of fit to the theoretical distribution is degraded. The Fisher z(r) test with equally weighted pairs is more reliable than any t(r) test based on conjectured degrees of freedom, but significance levels must be established by simulation if the number of pairs is small or redundancy is large. An approach to parametric analysis through nuisance parameters is discussed.