Three LOD score statistics are often used for genome-wide linkage analysis: the maximum LOD score, the LOD score statistic proposed by Kong and Cox, both based on the allele-sharing between affected sib pairs, and the maximization of the LOD score function of Morton on two genetic models and an heterogeneity parameter.Using only identity-by-descent sharing between affected sibs as linkage information, we studied the behavior of these three statistics under the null hypothesis in the rheumatoid arthritis simulated data (Genetic Analysis Workshop 15 Problem 3 - simulating model known). Distributions under the null hypothesis show that identical values of the statistics correspond to very different genome-wide p-values: comparison and interpretation of several linkage statistics cannot be done on the observed value. The Kong and Cox LOD score statistic had slightly better power to detect the HLA region involved in rheumatoid arthritis compared to the other methods. In a second step, we show that performing the analysis under a greater number of genetic models in the hope of better scanning the space of models, does not increase the power of detection.