Strain sensor is a crucial component in pavement response monitoring, and its measuring accuracy is vital to the evaluation and prediction of pavement performance. However, measurement variability and biases are unavoidable in nature due to the inherent granular characteristics of the asphalt mixture and the inclusion of the embedded strain sensor, respectively. In this study, a certain amount of 4-point bending beams, which were filled with random aggregates and asphalt mortar utilizing the finite element method, were constructed to represent the variability of the conventional dense asphalt mixture AC-13. Fiber Bragg grating sensor models of various lengths, anchor radii, and encapsulating moduli were then inserted into these bending beams to analyze the inclusion effect of the embedded strain sensor. The simulation results illustrated the diverse effects of the different geometries and moduli of embedded sensors on the stress and strain states of the asphalt mixture. From a purely theoretical perspective, a calibration equation was proposed between the theoretical value that represented the equivalent strain of the asphalt mixture and the measured value that was calculated from the sensor model. Multifactor variance analysis and multiple comparison procedure were applied to evaluate the measurement accuracy and to optimize the geometries and moduli of sensors. This research provides a basis for optimizing strain sensors employed in asphalt pavements and offers a novel insight toward the response measurement for granular materials.