The research outlines the ongoing development of a clinical prediction model to assist physicians in diagnosing pediatric asthma severity in the emergency department. It distinguishes between mild and moderate/severe asthma exacerbations using a limited amount of data from the patient’s history and clinical examination. As patient severity dictates time-sensitive management outlined in treatment guidelines, the model must provide an assessment very early in the management process (within 2 hours of arrival). Our model is based on retrospective chart data of emergency asthma patients assessed at an academic children’s hospital. The data is characterized by a relatively large number of attributes and a high ratio of missing values. From this data we have used machine learning techniques to develop decision models that may be easily interpreted by physicians (e.g. tree-based and rule-based). In order to improve the quality of the model we used external clinical knowledge (PRAM index) to partition the data set into different sets for classification. We have also applied contextual normalization to process values of clinical attributes depending on the patient’s age. Our experimental results demonstrate that the best decision model is a tree-based model developed from normalized cases.