Efficient management of a gas distribution network requires the ability to forecast demand for gas over various timescales. Gas demand varies from day to day for a number of reasons, many of which can be included in a model. We describe a project to construct a new model for forecasting gas demand in two neighbouring regions of the UK. In particular we include a novel approach to modelling the effects of public holidays. We fit the model using Bayesian inference which allows the use of expert prior beliefs, flexibility in aspects of the model, and the production of forecasts with associated uncertainty measures which reflect uncertainty in model parameters as well as “random” variation. The results of the work are now used by a gas distribution company.