Travel time reliability has received much attention from researchers and practitioners because it not only affects driver route choice behavior but also is used in the assessment of transportation system performance. Current state-of-the-art travel time reliability research assumes that roadway travel times follow a unimodal distribution. Specifically, a Weibull, exponential, lognormal, or normal distribution is used within the context of travel time reliability modeling. However, field observations demonstrate that roadway travel times are multimodal, especially during peak periods. This paper demonstrates that the multimodes observed in field data are a result of temporal variations in travel times both between and within days. This study evaluates the potential for generating multimodal travel times by using INTEGRATION software and investigates the underlying traffic states that result in these multimodes. The study validates the two-state model proposed in an earlier study and demonstrates the robustness of model parameter estimates under varying stochastic traffic congestion levels and sampling techniques.