This work addresses the problem of identification of modal properties, namely partial modeshapes, damping ratios, and natural frequencies, of a non-classically damped system under ambient excitation using a Bayesian probabilistic method. Ambient vibration tests eliminate the need for expensive dynamic experiments which take a lot of time, require permission, and cause problems to people who avail the benefits of the structure. Under operating conditions, naturally occurring vibrations can be measured and subsequent modal identification can be performed. For a Multiple Degree of Freedom (MDOF) system, the first few modes make the greatest contribution to the structural response. The operational excitation is assumed to be broadband enough to excite the most important modes of the structure so that the response can be effectively used for modal identification. In this work, a probabilistic approach based on Bayesian inference is presented to identify the complex modal parameters of a linear MDOF structure using output-only response data in the time domain. Computational obstacles in the estimation of probabilities are also overcome by using an approximate expansion for the likelihood function. The effectiveness of the approach is tested on simulated response data from a 2-DOF system.