Marine reserves (no-take zones) are widely recommended as conservation and fishery management tools. One potential benefit of marine reserves is that they can reduce fishing mortality. This can lead to increases in the abundance of spawners, providing insurance against recruitment failure and maintaining or enhancing yields in fished areas. This paper considers the factors that influence recovery following marine reserve protection, describes patterns of recovery in numbers and biomass, and suggests how recovery rates can be predicted. Population recovery is determined by initial population size, the intrinsic rate of population increase r, and the degree of compensation (increases in recruits per spawner as spawner abundance falls) or depensation (lower than expected recruitment at low abundance, Allee effect) in the spawner-recruit relationship. Within a reserve, theoretical recovery rates are further modified by metapopulation structure and the success of individual recruitment events. Recovery also depends on the extent of reductions in fishing mortality (F) as determined by the relationship between patterns of movement, migration, and density-dependent habitat use (buffer effect) in relation to the size, shape and location of the reserve. The effects of reductions in F on population abundance have been calculated using a variety of models that incorporate transfer rates between the reserve and fished areas, fishing mortality outside the reserve and life history parameters of the population. These models give useful indications of increases in production and biomass (as yield per recruit and spawners per recruit respectively) due to protection, but do not address recruitment. Many reserves are very small in relation to the geographical range of fish or invertebrate populations. In these reserves it may be impossible to distinguish recovery due to population growth from that due to redistribution. Mean rates of recovery can be predicted from r, but the methods are data intensive. This is ironic when marine reserves are often favoured for management or conservation in data-poor situations where conventional stock assessment is impossible. In these data-poor situations, it may be possible to predict recovery rates from very low population sizes by using maximum body size or age at maturity as simple correlates of the intrinsic rate of natural increase.