This paper reports on an experiment designed to investigate the effect of structural and linguistic variables on level of difficulty in solving arithmetic word problems. Identification of such variables is intended to assist curriculum writers in preparing exercises at a specified level of difficulty for students at various age levels. The study also considers the variables under varying time conditions and seeks to devise a coding system of the linguistic variables that would improve the accuracy of a linear regression model previously used in similar investigations. Details of the theory and methodology of the experiment are provided, and the results are discussed. The significance of the structural and linguistic variables is noted. (VM)