This paper reports on the results of a record check study evaluating measurement errors in the Survey of Income and Program Participation (SIPP).  Results indicate that, while misclassification errors in reports of participation level and change are rare, they sometimes result in important amounts of net response bias and in important amounts of attenuation of estimates of relationships.  We discuss some statistical strategies for understanding the response errors and introducing corrections, and we discuss behavioral strategies that could prevent errors from occurring in the first place.