Estimates of growth parameters are required for all length-based stock assessment methods. Most existing methods that estimate growth parameters from length-frequency data have important shortcomings. In this paper two new methods are subjected to Monte Carlo tests. The first method (Shepherd's Length Composition Analysis, SLCA) uses the information on the mean and distribution of length-at-age associated with cohorts, contained in length-frequency distributions. The second method (projection matrix) uses the information on growth rate contained in sequences of length-composition samples taken within a growing season. Results are presented for a range of length-frequency distributions under different levels of variation. If the order of magnitude of the parameters to be estimated is known a priori , the estimates are generally biased by less than 15%. The projection method, in particular, is robust to high levels of variation in length-at-age.