In this paper, we present a new analytical model that can give an accurate estimation of the blocking probabilities in wavelength-routed optical networks with heterogeneous traffic. By heterogeneous, we mean that each session offered to the network has its own traffic intensity and burstiness. In such cases, the blocking probability of a session is determined by the busy-wavelength distributions of the links seen at the arrival points of its calls. Thus, we first present two single-link models to estimate the arrival-point busy-wavelength distribution of a link with heterogeneous traffic: the full-population (FP) model and the reduced-population (RP) model. Both models are based on the BPP/M/W/W model, where the first two moments of an arbitrary session are matched by those of a birth-death process whose arrival rate linearly varies with the average number of busy wavelengths occupied by its own calls. We show that different sessions have different arrival-point busy-wavelength distributions depending on the burstiness of their traffic, i.e., a bursty session observes the link more congested than a smooth session. Next, we provide two extensions of the single-link models, the FP-full-load link-correlation model and the RP-reduced-load link-correlation model, to estimate the blocking probabilities of optical networks with heterogeneous traffic and sparse wavelength conversion. Both models employ the existing link-correlation models to take into account the occupied-wavelength-index correlation between two adjacent links. By comparing the results from the models with simulation results, we demonstrate that both models well approximate the blocking probabilities of individual sessions, as well as the network-wide blocking probability, for a wide range of traffic intensity, burstiness, and heterogeneity.