Typical statistical model building approaches neglect model selection uncertainty. Model selection uncertainty arises when a single "best" model is chosen from a set of models and analysis proceeds as if the selected model is the true model. This standard practice ignores that there may be many plausible alternative models, some of which may result in very different predictions. By failing to take into account the uncertainty in the selection of the model, inferences and predictions based on only the single selected model are overconfident. Bayesian model averaging (BMA) provides a means for accounting for model selection uncertainty in statistical inferences and predictions. BMA has been applied in a range of disciplines, and software now exists for implementing BMA using common classes of models, including linear regression, generalized linear models, and survival models. BMA also provides a framework for incorporating prior information about the relative plausibility of models within the model space. By averaging over several models, BMA can also improve predictive performance. The theory and use of BMA were discussed and an example application of BMA for predicting net radiation from more commonly measured and easily calculated climatological data were provided to illustrate the use of BMA in a practical context.