The Privacy Act of 1974 places restrictions on the Federal, state and local agencies' use of the Social Security account number as an identifier. For some agencies, compliance will involve changes in implementation of retrieval algorithms. This report describes methodology applicable to these changes in the more general context of the problem of retrieving individual records from files using non-unique identifiers. State-of-the-art retrieval techniques are discussed, a method for assigning reliability weights to various personal data elements is presented, file validation techniques for the error and omission rates of data items are suggested, and a retrieval probability model--designed to show likelihood of retrieval of a subject's record given a variety of populations, combinations of identifiers, and error/omission rates--is described. A methodology is developed for forming confidence factors from the established error/omission rates for combinations of non-unique identifiers that are candidates for use as retrieval keys. Use of these confidence factors as indices into the precision tables produced by the probability model is described. (Author)