The importance of forecasting brand sales has grown as markets become increasingly competitive and the cost of failing increases. Approximately two-thirds of new consumer product goods are discontinued within two years of launching. An accurate forecast of brand sales based on a robust measurement of consumer preferences can help marketing managers avoid costly failures before they get to market. However, the current approaches that measure consumer preferences suffer from a variety of limitations. These include: 1) difficulty interpreting claimed purchase intent data unless it has been calibrated to actual purchase data, 2) choice based conjoint (CBC) approaches that are too complex and costly inhibiting widespread application, and 3) models of historical sales data (Le. marketing mix models) that are not always viable due to the lack of data availability in many markets, and have limited application regarding decisions on launching new products or new marketing vehicles. In this thesis we examine how choice models can help address the limitations of current approaches that measure consumer preferences. Choice models, based in ii this research on either constant sum or CBC, are used to estimate the share of preference for a brand under various marketing conditions. A variety of aspects affecting the utility of choice models for forecasting purposes are explored in this thesis. We look at how context effects can be used to minimise response bias at the data collection stage, how the Dirichlet model can help estimate new product trial at the analysis stage, and how marketing mix models can be enhanced with data from choice models. Through the examination of these applications of choice models, we demonstrate how many of the limitations of the current methods can be overcome, which can help improve the decisions made by marketing managers. An underlying theme of this thesis is the importance of model validation, with many of the current methods lacking in this regard. The importance of external validity is examined, and the external validity of choice models (based on CBC and constant sum) are reviewed for use in a variety of applications. Understanding model accuracy, as determined through a validation exercise, is instructive to marketing managers as it informs them on how much confidence they should place in the model when making a decision. Understanding model validity is also critical for researchers as they seek ways to improve their models.