The formal basis for a data evaluation method is documented. This method has been successfully applied to a variety of practical problems, and details important to these problems are included. The method can be crudely described as a means for using lognormal a priori information in the standard least-squares algorithm. Emphasis is on the proper use of uncertainties and correlations, both in the data to be evaluated and in the final evaluation itself. Finally, important details and pitfalls common to least-squares methods are included.