We investigated to what extent a prediction of the 'ideal' hematocrit based on individual hemorheological profile with an equation of viscosity is relevant in trained athletes, and how the agreement between theoretical and actual values is modified by changes in training volume and performance. Elite soccer players (national level: 18-32 yr, weight 61-83 kg, body mass index 20.9-25.8 kg/m2) were seen twice at one year interval. Hemorheologic parameters were measured with the MT90 viscometer and the Myrenne aggregometer the theoretical bell-shaped curve of hematocrit/viscosity ratio as a function of hematocrit was reconstructed with Quemada's equation using actual plasma viscosity and red cell rigidity to predict hematocrit/viscosity at various hematocrit levels. RBC aggregation is correlated at baseline with fat mass (M1 = 0.552 p < 0.02) and changes in aggregation are related to changes in fat mass (M = 0.652, p < 0.05; M1 = 0.647, p < 0.05). Predicted and actual hematocrit are correlated (r = 0.644, p < 0.05) but exhibit discrepancies (mean difference -1% range [3.24 to 1.24]) and those discrepancies are inversely correlated to the level of predicted hematocrit (r = -0.912, p < 0.01), to systolic blood pressure (r = -0.626, p < 0.05), and to the overtraining score (r = -0.693, p < 0.05). After one year changes in hematocrit are a close reflect of the change in training volume (r = -0.877, p < 0.01) but are not correlated to fitness changes. Therefore in these athletes i) systemic hematocrit is close to its predicted 'ideal value", suggesting the accuracy of the prediction; ii) red cell aggregation is correlated to fat mass even in nonobese subjects; iii) hematocrit is lower than predicted by the model when markers of sympathetic tone (systolic blood pressure, overtraining score) are increased; iv) weekly training volume appears the main determinant of the reduction of hematocrit.