Prediction is more than testing established theory by examining whether the prediction matches the data. To show this, I examine the practices of a community of scientists, known as threaders, who are attempting to predict the final, folded structure of a protein from its primary structure, i.e., its amino acid sequence. These scientists employ a careful and deliberate methodology of prediction. A key feature of the methodology is calibration. They calibrate in order to construct better models. The construction leads to knowledge of how to construct or build an object. Thus, prediction serves a cognitive goal of model construction and not just model or theory testing. The kind of knowledge that results is relevantly different than theoretical knowledge.