For dynamic load identification， the identification of distributed load can be converted to that of the orthogonal polynomial coefficients by introducing the generalized orthogonal polynomial. Thus， the dynamic calibration technology is constructed to study the relationship between the system responses and the orthogonal polynomial coefficients， which is the most important factor affecting the accuracy of load identification. The traditional dynamic calibration method， that is simulation dynamic calibration， is applied based on the finite element simulation model， which has the model error that leads to the low identification accuracy of the dynamic load. Therefore， in this paper， the experimental dynamic calibration method is proposed based on Legendre orthogonal polynomials and Gauss-Legendre integral to reconstruct the distributed load in frequency-domain. In summary， the dynamic calibration is completed by measuring the frequency response function between the limited Gauss-points and the response points， which overcomes the limitation of the traditional calibration that cannot apply the orthogonal polynomial loads to solve the calibration matrix. Meanwhile， the influence of model errors on the accuracy of the calibration matrix is avoided to improve the accuracy of load identification. To verify the effectiveness and accuracy， simulation examples are studied to compare the proposed method with the traditional simulation dynamic calibration， and the influences of different Gauss-points on the identification accuracy and anti-noise performance are also discussed. Moreover， the experiment is also performed to further validate the feasibility and engineering applicability.