We investigate two parametric approaches and one non-parametric approach to estimating Internet users' value-of-time, an important characteristic of demand for Internet services. The advantages of these approaches are made clear and their limitations discussed. The models are tested with data generated from our simulation model of the Internet economy. Given the characteristics of the data, we investigate parametric count-data models first. While reasonably good results are obtained on all medium- and large-sized jobs, the method fails on small-sized ones. Second, we develop a nonparametric quasi-Bayesian update algorithm for retrieving the underlying distribution function of Internet users' value-of-time purely from observations on their choices. Compared with the parametric count-data models, good results are obtained in every case.