The spectral reflectance of the sea bottom plays an important role in radiative transfer signal of optically shallow water, affecting the spectral characteristics of water-leaving radiance in sea surface. Therefore, the precise information of the substrate is particularly significant in the study of the coastal remote sensing. In order to provide an accurate and convenient on-site information extraction of optically shallow water bottom, a set of sea bottom reflectance measurement system was designed, which is characterized by designing a reference whiteboard that can be freely stretched and rotated in order to eliminate the influence of the absorption attenuation of water between the probe and the target. And the design of the dual optical path for simultaneous measurement solved the interference of the spatial and temporal variability of water optical properties. Optical in-situ bottom reflectance of various substrates include corals, seagrass, sediment and beach collected in Sanya Coral Reefs Reserve during September 3—8, 2018 were used to study the feasibility of the system. As expected, the various types of substrate have distinct spectral separability. The spectral reflectance feature of the sand bottom and onshore beach are different over the range 580～700 nm, which suggests that the absorption and scattering of water and the presence of microalgae strongly affect the measurement of under water radiation. The difference between seagrass and coral is obvious that there is positive reflectance feature at 540～600 nm of seagrass, whereas the typical characteristic of coral is that there are three positive features around 575, 600 and 650 nm. In addition, the three carbonate substrates of coral, sand bottom and sandy beach have reflection peaks at 395, 430, 490 and 520 nm, and a small absorption peak at 485 and 585 nm, while seagrass has an absorption peak at 395, 430, 490 and 520 nm, a reflection peak at 485 and 585 nm. The above data results laid the foundation for the future extraction of benthic composition information, and also confirmed the reliability and real validity of the system design.