This study investigated the effects of 40-weeks training in swimming, cycling and running on performances in swimming, running and triathlon competitions in four elite triathletes. The training stimulus was calculated using the exercise heart rate. The level of performance was measured in running by a submaximal 30 min run, in swimming by a 5 x 400 m all-out test and subjectively in triathlon competitions. A mathematical model using one to three first order transfer functions linked actual and modelled performances by minimizing the residual sum of squares between them. The relationships between training and performances were significant in running (tau(1) = 20; tau(2) = 10; r = 0.74; p < 0.001) and in swimming (tau(1) = 31; r = 0.37; p = 0.03), supporting the principle of specificity of the training loads. Cross-transfer training effects were identified between cycling and running (tau(1 = )42; r = 0.56; p < 0.001), but not with swimming performances. In addition, the training loads completed in running were shown to have a major effect on performances in triathlon competition (tau(1 = )52; tau(2 = )4; r = 0.52; p < 0.001), indicating that running training is an essential part of triathlon performance. Swimming appears to be a highly specific activity, which does not gain nor provide benefits from/to other activities (i. e. cycling and running). The present study shows that cross-transfer training effects occur between cycling training and running performance in elite triathletes. A similar cross-training effect does not seem to occur for swimming performance.