In 1989, The Cornell Program on Dairy Markets and Policy collaborated with the Texas A&M Agricultural and Food Policy Center to form a National Institute for Livestock and Dairy Policy (NILDP). The Institute is a focal point for a neutral and objective analyses of the consequences of alternative government policies on the livestock, dairy, and poultry industries and the broader economics of livestock and dairy markets. Based on their respective strengths and emphases, Texas A&M is the lead institution on livestock and poultry sector analysis, and Cornell is the lead institution on dairy sector analysis. The Institute has been supported by a special research grant through the U. S. Department of Agriculture since 1989. The Dairy Farm Analysis Project (DFAP) is one particular effort in a larger set of objectives and core projects. In 1992, under the umbrella of the Dairy Farm Analysis Project, researchers from New York, Pennsylvania, Wisconsin, Michigan, and Ontario met to discuss the possibility of creating a pooled data set from the represented states. Raw data would not be collected under this project, but rather would be merged from individual state efforts already in place. From the four states and one province, a single data set was created which contains 2,200 individual farm level records with 92 basic and 15 calculated variables for the 1992 calendar year. This publication: describes the sources of the pooled data; discusses the representitiveness of the sample; provides a description of the definitions of variables; and summarizes the data with descriptive statistics. With this proj ect, we have shown that variables from different states I dairy farm record systems can be defined such that common variables can be obtained. We have developed a rich data set containing 1,818 farm records from four states and Ontario. While there are differences in dairy farm performance and profitability between states, the differences are more related to herd size differences than to other factors. In other words, farms of similar herd sizes are more like farms in other states of the same size, than to different size farms within the state. The pooled data set has shown that rates of production and profitabiltiy are higher on larger farms, even though operating cost of producing milk is higher. Labor efficiency on larger farms is significantly higher than on smaller farms. Larger farms have higher net worth, but also have higher debt to asset ratios and debt per cow. The most common herd size category in the data set is 40 -79 cows. This herd size is confronting high investments per cow, no advantage in debt per cow, and modest labor efficiencies in comparison to larger herd sizes. Their advantage is low operating costs, .primarily due to most of the labor being provided by the operator and family. However, the return to labor and management per operator is negative, as is return on equity with appreciation. This herd size, perhaps more than any other, will be struggling with high feed costs and the decision to expand in the future.