This paper introduces a new approach to constructing prior distributions for Bayesian models, focusing on penalizing the complexity of model components. The authors propose *Penalised Complexity* (PC) priors, which are designed to be informative yet interpretable and easy to elicit. These priors are defined using a user-defined scaling parameter and are formulated to be invariant to reparameterizations, connected to Jeffreys' priors, and robust to the choice of the flexibility parameter. The paper discusses the challenges of specifying priors in complex models and highlights the limitations of existing methods, such as reference priors and weakly informative priors. The authors provide theoretical and empirical results to demonstrate the effectiveness of PC priors in various scenarios, including hypothesis testing and model selection. The paper also includes a detailed construction of PC priors for univariate and multivariate parameters, with a focus on additive models and hierarchical models. The authors conclude by discussing the advantages of PC priors and their potential applications in practical Bayesian inference.This paper introduces a new approach to constructing prior distributions for Bayesian models, focusing on penalizing the complexity of model components. The authors propose *Penalised Complexity* (PC) priors, which are designed to be informative yet interpretable and easy to elicit. These priors are defined using a user-defined scaling parameter and are formulated to be invariant to reparameterizations, connected to Jeffreys' priors, and robust to the choice of the flexibility parameter. The paper discusses the challenges of specifying priors in complex models and highlights the limitations of existing methods, such as reference priors and weakly informative priors. The authors provide theoretical and empirical results to demonstrate the effectiveness of PC priors in various scenarios, including hypothesis testing and model selection. The paper also includes a detailed construction of PC priors for univariate and multivariate parameters, with a focus on additive models and hierarchical models. The authors conclude by discussing the advantages of PC priors and their potential applications in practical Bayesian inference.