The need to be able to design for a specific level of reliability and account for the inherent variations in the different design factors has been recognized in many areas of design. Basic stochastic design theory was applied to the structural subsystem of the Texas flexible pavement design system by Darter and Hudson, and this allowed the designer to determine the reliability of a specific design or to design for a specific level of reliability. So that this added feature of the system could be used, estimates of the variations associated with the different design factors were needed. A study was made to quantify the variations associated with the initial serviceability index of flexible pavements. The study was designed to obtain better estimates of expected average values and the variations of this design factor. A total of 21 newly constructed pavements were measured by using the surface dynamics profilometer. Most of the models used in the flexible pavement system are empirical equations based on data collected from test sections and in- service pavements. There is a certain lack of fit associated with the equations, and this lack of fit has to be estimated to determine the variance of a design. The lack of fit associated with the performance equation used in the flexible pavement system is believed to be quite large because of wide variations in traffic, environmental conditions, and materials in Texas. A method for determining the lack of fit of the performance equation is also presented.