This thesis develops statistical methodology for modelling the selectivity of fishing nets, using data from covered codend experiments of fishing trawls. First, the effects of subsampling an experimental catch instead of measuring the entire catch are investigated. Often the subsample is not taken at random. This leads to bias in the selectivity parameter estimates. Simulations show that the effects of non-random subsampling are minimised when equal proportions are sampled from the test and the control net. A model is developed for describing the selectivity of a net with a window panel inserted. This model quantifies the selectivity of both the codend and the window panel, which can then be combined. The model is used to investigate the selectivity of different window panels and their contribution to the combined selectivity. Traditionally, selectivity has been modelled as a fixed and random effects model, estimated in two stages. As an alternative, Markov chain Monte Carlo techniques are explored. A Bayesian selectivity model is formulated, and the effect of the prior distribution on the variance components is investigated. The posterior distribution is relatively insensitive to a prior distribution having variation of similar magnitude as the variation present in the data. For model selection, the p-value approach applied to the posterior marginal densities is more useful than the Bayes and pseudo-Bayes factors. The Bayesian selectivity model is extended to include variation between seasons and variation between trips. The new model is applied to a data set containing seasonal variation. Finally, a Bayeisan multi-species model is developed that accounts for dependencies between species. This gives more precise selectivity parameter estimates. It also reduces bias in the parameter estimates by accounting for mechanisms behind non-random missing data.