The connections between structural components, i.e., joints, are extremely difficult to model accurately by using a pure analytical approach. Alternatively, joint properties can be extracted from experimental data. A method for identifying (extracting) joint properties from the measured frequency response function (FRF) data is presented in this paper. One of the major difficulties in accurate identification of joint properties is that the identification results can be significantly affected by various errors in the measured data. Even if errors in the measured data are at an insignificantly low level (say, 5% in FRFs), their effects on the identification can be severe enough so that in some cases the predicted joint properties can be completely irrelevant to the true properties of joints. Apart from the accuracy of the data used for the identification, the accuracy of the identification also significantly depends upon the mathematical formulae used for the identification. Even with the same experimental data, the accuracies of the identification by different methods can be significantly different. Therefore, developing methods that are insensitive to measurement errors is of utmost importance for the identification of joint properties from experimental data. In this paper, techniques for improving the accuracy of joint identification are discussed. It is demonstrated that by using all available information effectively, the accuracy of the identification can be much improved. Both numerical and experimental examples are presented.