Circular data are data that can be expressed as directions on a unit circle, either directly when it concerns measurements from a circular scale, e.g. the compass or a circumplex measurement instrument, or indirectly when it concerns measurements of periodic patterns. Projected normal (PN) distributions allow for flexible modelling of circular data by introducing the assumption that circular data arise from a projection of a (latent) vector in bivariate space onto the circle. It allows for a smooth translation of existing models for linear data to the circular case. However, with respect to the interpretation of these models some difficulties arise. Making use of a Bayesian framework, we introduce new tools for PN regression and mixed-effects models that overcome these difficulties. Additionally, an R-package (bpnreg) was created that allows applied researchers to use the new interpretation tools and fit both Bayesian PN regression and mixed-effects models. By means of step-by-step analyses of example data we outline the use of the tools from the package and their usefulness to applied researchers.