Abstract: The real-time attitude information of agricultural small unmanned aerial vehicle(UAV) is a key factor to the decision and operation for variable rate program in precision agriculture. A real-time attitude estimation system of agricultural small UAV is designed here which consists of a microprocessor STM32, micro-electro mechanial system(MEMS) inertial sensors, and wireless transceiver module nRF24L01a. Detailed description and derivation of sensor calibration method and the multi-sensor fusion algorithm of attitude estimation based on quaternion derivation and the gradient descent algorithm are presented in the paper. Experimental results show that the sensor data acquisition and filtering consumes about 6.2ms, and the algorithm consumes about 0.96ms with the step size =0.8 in the 72MHz clock frequency and soft IIC(Inter-integrated circuit). The update frequency of attitude data up to 100Hz can meet real-time requirements. Statistics shows that the static mean absolute errors of pitch and roll are below 1.5 and the mean absolute errors of yaw are below 2.9. The mean absolute error of pitch, roll and yaw will be increased by 1~2 under the condition of micro-vibration of low frequency. It is indicated that the attitude estimation fusion algorithm and sensors calibration method are practical and effective which could provide accurate attitude data for precision flight control and variable operations implementation of agricultural small UAV.