Most fault-tolerant schemes for wireless sensor networks focus on power failures or crash faults. Little attention has been paid to the data inconsistency failures which occur when the binary contents of a data packet are changed during processing. In this event, a faulty node may produce incorrect data and transmit them to other sensors. Hence, erroneous results are propagated throughout the entire network and the sink may make inappropriate decisions as a result. Accordingly, this study proposes a mechanism which can both tolerate and locate data inconsistency failures in sensor networks. Node-disjoint paths and an automatic diagnosis scheme are utilized to identify the faulty sensor nodes. The proposed mechanism was implemented with the ns-2 simulator. The evaluation results demonstrate the ability of the mechanism to identify faulty nodes efficiently and with limited overheads. Using the scheme, more than 99.99% of the inconsistent data packets were successfully detected by the si...