Recreational anglers are known to seek different types of experiences in different settings. Such preference and behavioural diversity has important management implications. Research methods which assume only a limited degree of preference heterogeneity and impose rigid substitution patterns can impair understanding of individual’s unique decision process and lead to misguided management and policy recommendations. This paper reports results from an internet-based stated preference survey of recreational trout anglers in the North Canterbury Region of New Zealand. A mixed logit model, which simultaneously specifies random parameters plus error components, is used to capture the extent of random preference heterogeneity in systematic utility along with variance differences in unobserved utility at the alternative specific level. These error components are independent of the random parameters and fully relax the IID property. The model is further generalised to control for heterogeneity in the means and heteroscedasticity in the variances of the random parameters as well as error components using angler’s self reported skill level. The performance of the extended mixed logit-error component specification is evaluated against multinomial logit and latent class.