A method for developing statistically parallel tests based on the analysis of unique item variance was developed. A test population of 907 basic airmen trainees were required to estimate the angle at which an object in a photograph was viewed, selecting from eight possibilities. A FORTRAN program known as VARSEL was used to rank all the test items by unique variance. The item with the highest unique variance was put into a pool, then correlated against each of the remaining items. Next, the best weighted combination of items in the pool was correlated with each of the remaining items until all the items had been included in the pool, and were in order of uniqueness. Odd-numbered items were then allocated to the first test form, and even-numbered items to the second. The two forms were then compared statistically for reliability and validity, and it was found that the means were equal, the variances were equal, and there was equal correlation with a criterion. In order to demonstrate that the technique's success was not a function of the nature of the angle estimation items, a replication on a right-wrong scored attitude measure was done, with similar results. (BW)