Keywords: Planning, Weak Methods Systems: MOLGEN; also compares MOLGEN with several other systems: GPS, TEIRESIAS / MYCIN, HEARSAY like systems such as SU-X, SU-P, and the Hayes-Roth errand program in ). Summary: Presents a view of problem solvers as a control hierarchy and introduces MOLGEN as such a system for planning molecular genetics experiments. Argues that task oriented systems in which the interpreter controls task selection are really just monolithic "black box" systems since task selection does not use domain knowledge. Further rejects simple fixes to this problem. Instead proposes that knowledge about actions, goals, etc. be explicitly represented in the same language used to represent the domain knowledge: a meta-level. Argues for a multi-layered system for handling this sort of knowledge. Discusses general domain independent heuristics (e.g. "think on the end before you begin"). Presents MOLGEN as a three level system with the lowest level being domain space (the actual objects and actions in the real world), the medium level being the design space (the plan being developed), and the top level being the strategy space (which mediates between general heuristics and standard least-commitment planning). Presents these three layers in detail: the design space possesses a hierarchy of actions in the world such as cleaving DNA, etc; the design space posses a hierarchy of operators for manipulating plans such as computing goal differences, specializing abstract plans, etc.; and the strategy space has operators that suspend or resume least-commitment goals, make heuristic guesses, and do backtracking. Compares MOLGEN with a variety of other systems. Presents some open issues.