An autonomic problem determination system can adapt to changing environments, react to existing or new error condition and predict possible problems. In this report, we propose such a system using dynamic and adaptive multi-levels dictionaries and "Singular Value Decomposition techniques" (SVD). Compared to standard SVD, our system uses an iterative method that enables dynamic interaction between events and the current dictionaries with its entries being updated continuously to reflect relative importance of each event, thereby accelerating its convergence. The system captures knowledge in a hierarchical form for complex knowledge representation. It does not require a formal knowledge model or intensive training by examples. It is efficient with sufficient accuracy for autonomic problem determination.