Automated model selection (AMS) is a procedure to select the best model from a set of candidate models (multi-model inference). This approach can be very useful when the investigator is dealing with a large number of predictor variables to explain a subject of interest (dependent variable). The objective of this lecture is to present the concepts of AMS and illustrate how this procedure can be used in the development of empirical models in Animal Science. The attendees should have some experience in data analysis and empirical models. At the beginning, the hypothesis, objectives, and potential variables associated with the subject of study will be discussed. Then, key concepts in development of a meta-analytic data set and examples from the NANP website (https://animalnutrition.org) will be presented. Data should be assessed for biological coherence and outliers need to be removed. For AMS, a mixed model is fitted using all predictor variables that potentially affect the dependent variable (global model), then a set of sub-models are derived from the fixed terms of the global model. The parameters of those models are estimated by the maximum likelihood method. The AMS approach can use one or more information criterion, but the Akaike’s information criterion corrected for small sample size (AICc) is often adopted. All candidate models are ranked based on the lowest AICc to obtain model weights, then the best set of candidate models are selected. The variance inflation factor (VIF) is used to evaluate the correlation between predictors in the model, and biological coherency of the best models are evaluated. The final stage compares the best candidate models selected (often <10 models) using the root mean square error (RMSE) and the concordance correlation coefficient (CCC). AMS is useful for the development of prediction models when a large set of potential predictor variables are available.