The invention presents a method of identifying faults on a DSL line, typically intermittent faults arising from unstable joints in the DSL line. The method collects errored seconds data at the DSLAM and at the customer's premises equipment (CPE, typically a home hub or router). The error data collected at the DSLAM are termed near-end errors, and the error data collected at the CPE are termed far-end errors. The near-end and far-end data is then analyzed by applying regression analysis to determine if there is a correlation or match between the two sets of data. Matching data patterns are indicative of unstable or bad joints in the DSL line, and are typically intermittent and located near the customer's premises.