A method of detecting fraudulent or erroneous data from a transaction data set is provided. A first transaction record having a plurality of key values is selected from a transaction record database. One of the key values is selected from the selected transaction record. The transaction record database is then queried for transaction records having the selected key value. A second database is compiled of transaction records that contain the selected key value. At least one other key value is then selected from the originally selected transaction record, and the transaction record database is again queried for transaction records also having the second key value. The results of the fist, second and any subsequent queries are added to a second or a suspect transaction database. Then, using the uncovered records, the transaction database is queried for the use of key values common to the uncovered set of transactions to see if additional records are suspect. A list of queried key values is maintained to prevent the unnecessary or redundant use of the same query of the transactions. Risk coefficients of levels of fraud or mistake are assigned to the transaction records which have one or more common key values to the records uncovered as a consequence of the database queries.