This paper discusses the overall design of the PIXIE Intelligent Tutoring System, and more specifically a series of recent enhancements. The original system has been implemented to involve three separate phases: the offline, or model generation phase; the online, or tutoring, phase; and the analysis phase. The offline phase, which is completed prior to any interaction with a student, involves the construction of a set of student models for a given domain. The online phase involves the tutorial interaction with a student, consisting of both diagnosis and remediation of errors. During the post-interaction analysis phase, undiagnosed errors are examined and, if consistent, added to the existing domain knowledge base. Four recent enhancements to the system are discussed, each arising from a shortcoming that was noted in the system as a result of student trials. Two of these additions to the system involve the diagnosis of errors, and two involve the remediation of errors. (Author/TW)