A cost of reproduction in terms of reduced future performance underlies all life-history models, yet costs have been difficult to detect in short-term experiments with long-lived plants. The likelihood of detecting costs should depend on the range of variation in reproductive effort that can be induced, and also on the shape of the cost function across this range, which should be affected by resource availability. Here, we experimentally examined the effects of both reduced and increased fruit production in two populations of the long-lived orchid Gymnadenia conopsea located at sites that differ in length of the growing season. Plants that were prevented from fruiting produced more flowers in the population with a longer growing season, had higher survival in the other population, and grew larger compared to control plants in both populations. Fruit production was pollen-limited in both populations, and increased reproductive investment after supplemental hand-pollination was associated with reduced fecundity the following year. The results demonstrate that the shape of the cost function varies among fitness components, and that costs can be differentially expressed in different populations. They are consistent with the hypothesis that differences in temporal overlap between allocation to reproduction and other functions will induce among-population variation in reproductive costs.