We review the statistical issues arising in modeling univariate extremes of a random sample, with emphasis on heavy-tailed Pareto-type distributions. In the last three decades, the semi-parametric approach based on probabilistic asymptotic results in extreme value theory has strongly developed. This approach was initiated with the estimator of the tail index proposed by Hill (1975). We present the basic ideas behind the different classes of estimators that have appeared in the literature. We emphasize methods that facilitate practical use. We illustrate the methods presented using a dataset of weekly returns of a European bank.