Pavements need constant rehabilitation when they deteriorate with time and approach the end of their expected service lives. Overlay is the most prevalent treatment that restores its desirable condition and extends its life span of serviceability, especially for roads subjected to moderate and heavy traffic. Overlay composite design remains a major challenge due to difficulties in characterizing the complex behavior and assessing the existing condition of a combination of asphalt concrete (AC) and Portland cement concrete (PCC) layers over a soil subgrade. Deflection based design using falling weight deflectometer (FWD) deflection data offers an effective approach for overlay thickness design for composite pavements. It utilizes the deflection measurements of the pavement surface which can be used to back-calculate the subgrade and overlay composite properties and allows one to estimate the structural capacity of the existing pavement. However, the prevailing deflection based design procedure generally treats the AC and PCC as a single layer during the back-calculation and, as a result, frequently leads to less than satisfactory, usually over-conservative, design for overlay composites. The principal objective of this research is to develop improved FWD deflection based design strategies for overlay composite pavements. It is proposed that a three-layer linear elastic model be used for back-calculation of the moduli of all three layers: subgrade, PCC and AC. The structural capacity of the existing pavement is estimated using pavement surface deflections measured by FWD, the most commonly used pavement non-destructive testing (NDT) device. In the present study actual FWD deflection data for eleven construction projects are used to back-calculate the moduli of three layers. The three-layer model allows the composite pavement structure to be modeled more accurately. The elastic moduli of the asphalt concrete layer and the underlying Portland cement concrete can both be back-calculated, instead of combining them into one. The results show that the three-layer model produces higher effective thickness than the two-layer model for the same pavement structure, thereby reducing the required overlay thickness. However, there are a number of factors that can strongly influence the final overlay design thickness. The effects of computational error tolerances in back-calculation, temperature at FWD testing and variations in FWD deflection data are found significant and may cause unreliable design results and hence, two strategies to avoid excessively large or small back-calculated moduli are also explored: imposing moduli bounds and relaxing the precision convergence; they have been found very effective in mitigating the effect of large variations in deflection data. The statistical variations observed in the overlay design are also evaluated and two models are explored to improve the overall design procedure from the statistical perspective: Monte Carlo method and Point Estimation method. The effective thicknesses of existing pavement computed from reliability analysis are similar to those obtained from the proposed design method. This demonstrates the validity of the proposed design method and also the applicability of reliability based design in case the statistical parameters are available or can be obtained from engineering judgment.