Summary Many perennial plants experience prolonged dormancy, meaning they do not grow above‐ground for one or several growing seasons. When plants disappear (fail to sprout) and have not been recorded to re‐emerge, they either have died or are alive and dormant. In demographic studies of such species, researchers have been forced to make assumptions about death versus dormancy. Little is known about the consequences of these assumptions for predictions from the population models used in the studies. Here, we define survival of dormancy‐prone plants in three distinct ways: Separate, Instant and Slow death. In the Separate death model, plants are assumed to die either in emergent or in dormant stages. Death can also be described as a process of disappearance followed by dying, either in the first year below‐ground (Instant death) or with constant mortality rate in any given dormant year (Slow death). Using simulated data with known parameter values, we test whether survival and re‐emergence rates are confounded in these life‐history models. Using a general model and models for two orchid species (Epipactis atrorubens and Isotria medeoloides), we test how different assumptions about dormancy (Instant and Slow death) affect predictions about population dynamics: population growth rate, generation time, relative reproductive values, life expectancies, and sensitivity and elasticity of population growth rate to stage transitions and vital rates. Our results confirm that survival and re‐emergence rates of dormancy‐prone plant species are difficult to estimate; parameters were separable only with the assumptions of Instant and Slow death. Both theoretical and empirical analyses show that the predictions of population growth rate and generation time do not depend on the assumptions made about the fate of the plants after their disappearance. Synthesis. Assumptions about dormancy affect some, but not all predictions, about plant population dynamics. In the studies of species with unobservable stages, assumptions about dormancy should be carefully defined and possible consequences of these assumptions for the predictions of the population model should be evaluated. Nonetheless, if we are only interested in overall population viability, ad hoc models of prolonged dormancy are sufficient as a first step.