A concurrent program fault can propagate both within a single task (thread of control) and between tasks, making fault localization difficult. We propose a "two-dimensional" approach and supporting techniques for integrating analysis of task interactions, inter- and intra-task data flow, and conventional sequential debugging of individual tasks. We define augmented concurrent dynamic slice, a variant of dynamic slicing, which balances the cost and accuracy for concurrent program debugging and permits adjustment of that balance and focus on small parts of large complex systems. We also describe the design and implementation of prototype tools which add concurrent slicing capability to an existing debugger.