A nonparametric model for estimating mesh size selectivity is presente
d based on a multinomial likelihood for sampling from a mixture of dis
crete size classes. No functional form for the selectivity curve is as
sumed when comparing the relative selectivity of standard and experime
ntal mesh codends. Auxiliary information from the size distribution of
fish gilled in the experimental mesh codend is used to identify when
the selectivity approaches 1. Analysis is illustrated for Dover sole (
Microstomus pacificus) caught in an alternate haul study. Parametric m
ethods of estimating selective curves based on logistic functions show
ed tendencies for systematic bias in the tail regions as well. as poss
ible errors in the mid-range (approximately 50%) of selectivity.