In a recent study of the kinematics of the drive phase of the rowing stroke, Lamb (1989) provided detailed evidence that ergometer performance simulates on-the-water performance closely. In the present experiment, Lamb's analysis was extended in an investigation of the timing of the complete cycle of the rowing action of 5 rowers under each of those performance conditions. The authors followed Beek's (1992) suggestion that the first task in the analysis of timing in skilled movement is to specify the sources of variance and invariance in each particular task by identifying the major temporal constraints and the key relative timing variables. In addition, the possibility that some simple mathematical relationship (e.g., Schmidt, 1985) might describe the relative timing between the stroke and recovery phases of the rowing action when performed at different speeds was investigated. Both an absolute and a relative variability criterion were used in assessing and comparing timing variability over 4 speeds of rowing and between on-water and ergometer rowing in 5 elite male subjects. Criteria outlined by Gentner (1987) were used in assessing relative timing between stroke and recovery. The results indicated that variability decreases dramatically as a function of increased rowing rate; however, when variability is expressed as a function of movement duration, those decreases appear much less dramatic. Overall variability of the rowing cycle was caused principally by variability in the recovery phase, whereas the stroke phase was relatively invariant under both rowing conditions. The changes in the relative timing of the rowing stroke across the 4 speeds studied followed a simple mathematical rule, best described as linear increments in the stroke proportion of the total rowing cycle with increases in rowing rate. Moreover, those changes were similar across the 2 rowing conditions. The present results are discussed in light of findings from other forms of propulsion, such as walking, running, and stair climbing, in which the movement constraints are quite different.