Several case-control association studies have raised the possibility that the A allele of a -1438 G/A polymorphism in the type 2A serotonin receptor (HTR2A) gene may be a risk factor for anorexia nervosa. However the absence of linkage and the existence of negative association studies raise the possibility of false positive findings, resulting from population stratification or lack of statistical power. To address this controversy we recruited a sample of 316 patients with anorexia nervosa from six European centres, and utilised a family-based transmission disequilibrium (TDT) approach to analyse the HTR2A-1438 G/A polymorphism. Age at onset and minimal BMI were also taken into consideration in order to detect clinical heterogeneity or a quantitative trait effect. The TDT approach showed that the A allele was transmitted 133 times and not transmitted 148 times (McNemar chi(2) = 0.29, df = 1, P = 0.59). Also, the haplotype-based haplotype relative risk method showed no evidence for association of the A allele, in samples from each centre (chi(2) < 2.15, df = 1, P > 0.14) and in the total sample (chi(2) = 0.55, df = 1; P = 0.46). Furthermore, we found no evidence for heterogeneity of the A allele frequency between samples (chi(2) = 2.54, df = 4, P = 0.64), either according to minimal-BMI (F1/242 = 2.14, P = 0.45) or age at onset (F1/224 = 2.39; P = 0.12). QTL-TDT analyses also showed no direct role of the A allele on these traits. We thus found no evidence for a significant role of the 5-HT(2A) gene in anorexia nervosa. Previous results may have been exposed to stratification bias (which we controlled by the TDT method) and/or the risk of type 1 error (from which we were less exposed because of the sample size).