This paper describes a range of computer assisted learning software models--linear, unstructured, and ideal--and discusses issues such as control, interactivity, and ease-of-programming. It also introduces a "compromise model" used for a package currently under development at the Hong Kong Polytechnic University, which is intended to teach students principles of molecular biology, incorporating three activities--concepts, practice, and assessment. Difficulties involved in trying to build intelligence into the practice sectors is also discussed, particularly the irony of designing software which can actually disadvantage good students. Eight figures present: an electronic book; unstructured hypermedia software; a cartoon of "the intelligent tutor"; a mild constructivist model made up of three-part units; a sub-menu of the molecular biology project; an example content page; an example activity; logic depicting possible feedback; and a more incremental and fair routine. (Contains 11 references.) (Author/DLS)