A detailed exploration of all the polymorphisms in candidate genes is required to better characterize the relationship between gene variability and complex traits. We propose a novel strategy for investigating the association between a highly polymorphic gene and a phenotype, by combining a multilocus genotype analysis and an haplotype analysis. For the multilocus genotype analysis, a data mining tool--termed DICE (Detection of Informative Combined Effects)--was developed to identify the best subset of polymorphisms that are associated--individually or in combination--with the phenotype. For the haplotype analysis, we used our recently developed method of haplotype-phenotype association to determine the most informative and parsimonious haplotype model fitting the data. We illustrate this strategy by investigating the association between twelve polymorphisms of the APOB gene and plasma apoB levels in 1442 European subjects. After exploring all main effects and interactions between polymorphisms, DICE identified the N4311S polymorphism as the most informative polymorphism in relation to apoB levels. Haplotype analysis led to the same conclusion. Additionally, DICE identified the E4154K (EcoRI) and the T2488T (XbaI) polymorphisms as potentially interesting. This selection was not modified by inclusion of the common APOE polymorphism in the analysis.