Concerned with the influence of multipath, this paper proposes a spatially differencing smoothing technique for the low-elevation estimation in the bistatic MIMO radar under the spatially colored noise. Firstly, the multipath environment for a low-elevation target in the bistatic MIMO radar is modeled, by considering the specular reflection of the transmitter and receiver. The diffuse reflection is assumed to be the spatially colored noise. Then, the covariance matrix differencing is used to eliminate the unknown noise component, and the transmitting array and receiving array are spatially smoothed to decorrelate the multipath signals, which does the spatially differencing smoothing operation. Finally, the Direction of Departures (DODs) and Direction of Arrivals (DOAs) are estimated by unitary Estimation of Signal Parameters using Rotational Invariance Techniques (ESPRIT) algorithm. This paper also points to the rank deficiency problem of the spatially differencing smoothed covariance matrix in a special case, and modifies the spatially differencing smoothing method correspondingly. The proposed methods require a small number of antenna elements, fit for general unknown noise fields and low SNR environment, and solve the angle-merging problem in joint DOD and DOA estimation. The simulation results demonstrate the effectiveness of the proposed method.