The cost estimates of highway pavement improvements are a vital input for project feasibility analysis, appraisal and evaluation, life-cycle costing, and long-term programming and budgeting. Common types of highway improvement projects include resurfacing and reconstruction. In the current environment of declining budgets, increasing focus on agency transparency and accountability, and increased need for optimal resource allocation practices across highway pavement improvement projects, the need for reliable cost estimates is greater than ever before. The traditional method of costing uses simple averaging (dollars per lane-mile). For a given thickness of laid pavement, this average cost concept is simple to use, tractable, and easily understood; however, it can be misleading because it fails to account for scale economies associated with project length (miles) and width (number of lanes) and can therefore lead to over/underestimation of actual project costs and serious bias in cost estimation (and consequently in evaluation), and therefore can jeopardize the integrity of the intended application of the cost estimate. As a related side issue, pavement managers are interested in knowing the sensitivity of the unit cost to changes in the project dimension, that is, length or width (number of lanes), and therefore avoid the pitfalls of the average cost concept. To address this issue, past researchers have developed cost estimation models that use the cost per lane mile as the response variable. This paper shows that to adequately characterize the scale economies of highway pavement improvements, it is preferable to use the total cost as the response variable and then differentiate the resulting with respect to the output variable. The analysis was carried out using a variety of linear and non-linear econometric specifications, and the best functional form was selected on the basis of the capability of the functional form to exhibit non-constant returns to scale across both dimensions. The data used were from actual contract costs of highway projects implemented in the 1994-2011 period at a Midwestern state in the USA. It is shown that the developed random parameters cost functions, in characterizing the sensitivity of pavement improvement project costs across each of the two dimensions, present a far more reliable alternative to the use of average cost for estimating the cost of these project types. Also, the paper shows that it is better to identify scale economies using the total cost functions instead of the unit cost functions.