The present study was performed to determine how equilibrated fluctuations in hydration affected the validity of bioelectrical impedance analysis (BIA) for body composition assessment. Total body water (TBW) expansion was induced by a 4 d endurance trial and the subsequent water loss was obtained over the recovery period. Twelve healthy men exercised on a cycle and treadmill alternately for 5 h/d over 4 d at moderate intensity. TBW, fat mass (FM) and fat-free mass (FFM) were assessed 3 d before the trial (control), and on the first and eighth day of recovery (R1 and R8 respectively). TBW was evaluated by 2H dilution (TBW2H) as a reference method and by BIA (TBWBIA) at 100 kHz at the same time. TBW2H increased significantly between the control day and R1 by 1·87 (sd 1·11) litres (P=0·005) and TBWBIA by 1·38 (sd 1·56) litres (P=0·009). Both values returned to the control level on R8. For each period, TBW2H and TBWBIA did not differ significantly and were correlated (r2 0·85, P=0·0004 for the control day; r2 0·63, P=0·03 for R1; r2 0·75, P=0·02 for R8). Plasma Na concentration and osmolality did not differ between the control day, R1 and R8. FFM gain (1208 (sd 1983) g) and FM loss (−1168 (sd 906) g) between the control day and R1 were followed by a FFM decrease (−624 (sd 1281) g) and a FM increase (860 (sd 1212) g) between R1 and R8. As expected, these FFM and FM changes were significantly correlated with TBW variations. The present results provide evidence that BIA may be a useful method for estimating TBW when fluid shifts are equilibrated and electrolyte concentrations are unchanged. However, it is not a valid technique for assessing FM and FFM under these conditions.