Kinematic calibration is a common method for enhancing the accuracy of articulated arm coordinate measuring machines （AACMMs）. However， the residual errors after calibration can affect its measurement accuracy and stability. In this study， we propose a residual error compensation method based on a compound calibration and extreme learning machine to improve the measurement accuracy of AACMMs. First， we establish the kinematic parameter identification model based on the kinematic modeling of AACMM. Furthermore， we conduct angle parameter identification， length parameter identification， and length parameter scaling to complete the compound kinematic calibration. Subsequently， we construct the measurement configuration with the measurement angle， elevation angle， distance， and rotation angle as variables to analyze the residual error map. The proposed residual error compensation method is based on an extreme learning machine owing to the strong nonlinear relationship between the measurement configuration variables and the residual errors. We verify the validity of the proposed method through experiments. The results show that the maximum value of the single point measurement error of the AACMM decreases from 0.061 mm to 0.044 mm， the mean value of measurement error decreases from 0.023 mm to 0.017 mm， and the standard deviation of measurement error decreases from 0.011 mm to 0.007 mm after residual correction. Furthermore， the maximum length measurement error decreases from 0.137 mm to 0.074 mm， the mean measurement error decreases from 0.033 mm to 0.021 mm， and the standard deviation of measurement error decreases from 0.037 mm to 0.019 mm.