Bayesian network structure discovery is a form of graphical modelling based on machine learning that facilitates interpretation of complex biological systems. This multivariate approach was applied to outline the inter-relationships between information collected in an animal welfare control programme for laying hens in Sweden. The resulting directed acyclic graph identified the housing and management factors associated, or not, with the animal welfare indicators. Type of housing system, farm tidiness, management routines for manure removal and barn infrastructures were the key drivers of animal welfare. The main advantage of this approach was the holistic view, accounting for mutual dependence of all variables and highlighting both direct and indirect pathways to reach the same welfare goals. This opens a whole range of future work exploring which of the available pathways represent the most costeffective intervention options, amongst other applications.