Cognitive science research offers hope for the development of innovative science teaching strategies that facilitate the development of optimally interconnected procedural and declarative knowledge networks. Improving students' neural networks should improve their abilities to think critically, reason logically, learn more efficiently, and solve more complex problems. The primary purpose of this research study was to reveal the intricacies of students' knowledge networks by constructing a cognitive network model associated with one particularly important problem-solving skill which is prediction. The central role of prediction in learning, problem solving, and scientific progress demands that its cognitive mechanisms be investigated and that teaching and learning strategies be developed with prediction as the focus. Fourteen prediction problem sheets, involving six different conceptual systems in biology and three different problem formats were administered to pre-service college students at a north central U.S. university enrolled in secondary or elementary methods classes. Each problem sheet required students to make a prediction, either in writing or on a graph, and then to explain their reasoning in writing. Qualitative data analysis followed a series of 4 steps, with each step building from the previous step: (1) identification of cognitive behavioral categories associated with successful and unsuccessful predictors; (2) construction of sequential behavioral transcripts of both successful and unsuccessful predictors; (3) identification of cognitive scripts for successful and unsuccessful predictors with attention to procedural and declarative knowledge; and (4) development of a cognitive-network model of prediction problem solving. An Explicit Prediction Teaching Strategy (EXPRTS), developed from the model, encourages students to build a predictive framework of successful cognitive scripts. The mechanism of successful prediction problem solving was shown to be dependent upon a complex interplay between procedural and declarative knowledge as illustrated by a cognitive network model. This research represents one significant step toward developing problem-solving teaching strategies that take more advantage of the tremendous potential of "prediction power" in the science classroom. (KR)