THE PURPOSE of this study is to contribute to the theoretical and empirical literature concerning the term structure of interest rates. The analysis shows that, when the assumption of accurate interest-rate forecasting is dropped, lenders and borrowers can be to develop a variety of portfolio preferences. These preferences are determined by the vector of real and monetary forces in the economic system. Factors influencing investor preferences include such elements as: trends in the investor's cash inflows and outflows, liabilities, assets, income, and various security-price variances. To the extent that each investor faces a set of forces peculiar to him, we expect him to develop a desired portfolio pattern peculiar to him. For example, the assets and liabilities of a commercial bank differ in nature from those of a life insurance company and, as a result, the maturity preferences of the two normally differ. This conclusion is basic to the market-segmentation theory and ignored in the purer, abstract forms of the expectations theory. In attempting to determine the empirical significance of the segmentation and expectations theories, elements of these theories are tested against data on interest rates and on the stock of government bonds outstanding. To test the assertion of the expectations theory that maturity composition of securities' demands are determined primarily by current and forward interest rates, variables reflecting the maturity structure of the U.S. government securities portfolios of seven major investor groups are regressed separately on several current and two types of expected interest rates. The seven investor groups consist of commercial banks, life insurance companies, fire, marine and casualty insurance companies, savings and loan companies, nonfinancial corporations, mutual savings banks, and other non-governmental investors. The data are monthly observations over several overlapping time periods, especially the 19461966, 1955-1966, 1960-1966 periods. The two types of expected short rates involved in the regressions are forward one-year yields imputed from current yield curves and forward three-month bill rates defined to be weighted averages of past bill rates. Because the signs of the regression coefficients frequently are not those the theory leads one to expect and coefficients of determination are usually small, the data do not support these elements of the expectations theory. In view of the weight given to Meiselman's findings suggesting that changing interest-rate patterns in part reflect rate-forecast adjustments made in response to learned errors in prior forecasts, an alternative formulation embodying an errorlearning model is also developed and tested. It is hypothesized that changes in the maturity structure of investors' portfolios are functions of learned forecasting errors. Again, unexpected signs found for the regression coefficients and small coefficients of determination weigh against the expectations hypothesis. The market-segmentation theory maintains that the term structure of interest rates reflects the relative demands of different investors with a variety of relatively inflexible portfolio preferences. This suggests that if the total net demands for securities by in-