As part of a project to develop an intelligent computer tutor for basic algebra, we have been investigating task sequencing. In this paper we present an approach to task sequencing that is based on a component-skills view of intelligence and learning. We postulate that tutors use inferences about past and present student performance to determine a current skill set that will be the new target for learning. The skill set is then used as a basis for generating tasks that should elicit those skills. Current skill sets are modified slowly over time so that lessons appear coherent and well-planned. We first describe the approach at a general level, where it can be viewed as a cognitive model of human task sequencing. Then we discuss the implementation of the model in our intelligent algebra tutoring system.