Conventionally, a single near-infrared (NIR) channel, like 3.75 microns, has been used to retrieve the cloud droplet effective radius (DER) information. However, due to the large absorption by cloud droplets at this band, the DER inferred could only represent a thin layer near cloud top, which is inadequate to represent the entire DER vertical profile. Since numerous in-situ observations have shown that low-level stratiform cloud DER profile often increases monotonically with height, additional information is thus required in order to properly determine the vertical variability of the DER profile. In this study, an efficient retrieval method is presented for inferring the cloud DER vertical profile by utilizing bi-spectral NIR split-window measurements at 1.65 and 3.75 microns. The fundamental of the retrieval method lies in different path lengths of photon transport within cloud between the two wavelengths. Photons can travel significantly deeper inside a cloud at 1.65 microns than at 3.75 microns. As such, an independent DER retrieval from the 1.65-micron channel conveys DER information at a larger depth within cloud than that inferred from the 3.75-micron channel. The proposed bi-spectral retrieval method assumes that both the cloud liquid water content and DER vary linearly with height, which should be valid for low-level non-precipitation clouds formed by an adiabatic or pseudo-adiabatic process. The method adopts a lookup table technique to search for an optimal linear DER profile that render the top-of-atmosphere reflectances to agree with satellite measurements at both 1.65 and 3.75 microns. The retrieval method is limited to uniform, low-level, single-layered water clouds.