<p>This article tackles the problem of uncertainty quantification and calibration in human trajectory prediction (HTP) tasks. Leveraging precise and meaningful uncertainty quantification is at the heart of risk-aware decision making in many areas relying on estimates provided by machine learning models, for example in autonomous driving. After reviewing the different ways to handle uncertainty representation in HTP, typically cast as a regression problem, we propose and compare three uncertainty calibration methods designed to be used as <i>post-processing steps</i> to <i>any</i> existing stochastic HTP method. Moreover, we present an importance sampling scheme to reflect the calibration effect on the same samples produced by the uncalibrated predictive model. We evaluate and compare these calibration methods combined with different predictive models on standard HTP benchmarking datasets. Our code is available at <a href="https://github.com/cimat-ris/trajpred-unc">https://github.com/cimat-ris/trajpred-unc</a></p>