Quadcopters or quadrotors have always desired to fly smoothly and stay on their path in order to enhance their application. This better performance can be achieved depending on the accuracy of the data. However, relying purely on sensor data cannot be accepted due to the inaccuracy of measurement, thus state estimation to filter noise out is important. This paper is focused on performance evaluation on quadrotors attitude estimation using Unscented Kalman filter (UKF) by comparison with quadrotors attitude computed from the mathematical model. The UKF is an algorithm dealing with noise filters that can be used for state estimation such as attitude and bias of sensors. UKF is divided into two steps which are Measurement Update, and Time Update. However, the algorithm is initialized by determining initially on the mean and the covariance. Then, the measurement update algorithm uses the accelerometer sensor data and magnetometer sensor data pose the next time update. Gaussian with covariance (Q and R) of the UKF algorithm are determined in this paper. Quaternion is used to describe state equations which are based on the kinematic model. The input of the state equation is taken from sensors of the Pixhawk 4 controller which are gyrometers (data of angular velocity). In addition, the output equation is based on accelerometer modeling and magnetometer modeling including data from accelerometer sensors and magnetometer in the Pixhawk 4 controller. MATLAB & Simulink have been used for this experiment and Pixhawk Controller hardware is used as the flight controller. The result of attitude estimation expressed about component of quaternions and bias of sensor from Pixhawk 4. The graph is shown the performance of attitude quadrotors and bias of sensor.