We describe a forward-time haploid reproduction model with a constant population size that includes life history characteristics common to many marine organisms. We develop coalescent approximations for sample gene genealogies under this model and use these to predict patterns of genetic variation. Depending on the behavior of the underlying parameters of the model, the approximations are coalescent processes with simultaneous multiple mergers or Kingman's coalescent. Using simulations, we apply our model to data from the Pacific oyster and show that our model predicts the observed data very well. We also show that a fact which holds for Kingman's coalescent and also for general coalescent trees--that the most-frequent allele at a biallelic locus is likely to be the ancestral allele--is not true for our model. Our work suggests that the power to detect a "sweepstakes effect" in a sample of DNA sequences from marine organisms depends on the sample size.