Mark-recapture techniques are used for studies of animal populations. With only three sampling occasions, both Bailey's triple-catch (BTC) and Jolly—Seber's (J—S) stochastic method can be applied. As marking and handling of fragile organisms may harm them, and thereby affect their chances of being recaptured, handling should be minimized. This can be achieved by taking a subsample before the main sample at the second sampling occasion. Individuals in the main sample are marked and released, whereas those in the subsample are only used for identifying recaptures. Monte-Carlo simulation was used to compare the subsampling method with the ordinary mark-recapture methods. Model-generated populations were sampled with and without subsampling to provide estimates of population size, loss, and dilution rates. The estimated parameters were compared with their true values to identify biases associated with the sampling methods, using 81 different combinations of population size, dilution rate, loss rate, and sampling effort. Each combination was replicated 1,000 times. In no cases did subsampling perform more poorly than the ordinary methods. J—S was slightly more accurate than BTC to estimate the population size, but only when sampling effort was high. The relative biases associated with estimates of dilution and loss rates were substantial, but declined with increasing population size and sampling effort. Confidence limits for the population parameters generally were reliable and tended to be conservative. We therefore conclude that ordinary mark-recapture methods can be supplemented with subsampling without sacrificing accuracy. Subsampling is especially advantageous in cases where marks are difficult to observe under field conditions.