We compare the powers of three methods for the QTL analysis of non-nor
mally distributed traits. We describe the nonparametric and the logist
ic regression approaches recently proposed in the literature and study
the properties of the standard regression interval mapping method whe
n the trait is not normally distributed. It is shown that the standard
approach is robust against nonnormality and behaves quite well for bo
th continuous and discrete characters. The loss of power compared with
the nonparametric or the logistic approach is generally minor. Moreov
er, the least squares estimation procedure of the regression interval
mapping is not affected by departure from normality. The use of other
approaches could be restricted to extreme cases where the trait distri
bution is very skewed.