We propose a definition of bias founded on the concept of the study base. The definition is unrelated to the study design, thereby permitting a clear comparison of validity among different study types. Also, it facilitates assessment of bias because it allows bias to be subdivided under mutually exclusive categories. In relation to the study base, the epidemiologic study can be viewed as involving three distinct steps. First, the person-time (study base) is specified for the study. Second, the investigators collect data on disease events and person-time among the exposed and unexposed from the study base. Third, they calculate an estimate of the effect of the exposure using the data. Categories pertaining to each of these three steps will create mutually exclusive categories of bias. We suggest the following terminology: the first step generates confounding, the second results in misclassification and misrepresentation, and the third produces analysis deviation. The total bias is the sum of the confounding, misclassification, misrepresentation, and analysis deviation. We illustrate our proposed epidemiologic theory with examples from various biological settings.