Multiple degree of freedom (MDOF) algorithms are the dominant methods for extracting modal parameters from measured data. These methods are founded on the notion that because the response of a linear dynamic system is the sum of many modal contributions, the extraction technique must deal with all of the modal parameters in a simultaneous fashion. The Mode Isolation Algorithm (MIA) described here is a frequency domain formulation that takes an alternative viewpoint. It extracts the modal parameters of each mode in an iterative search, and then refines the estimation of each mode by isolating its effect from the other modal contributions. The first iteration estimates modes in a hierarchy of their dominance. As each mode is estimated, its contribution is subtracted from the data set, until all that remains is noise. The second and subsequent iterations subtract the current estimates for all other modes to identify the properties of the mode under consideration. The various operations are described in detail, and then illustrated using data from a four-degree-of-freedom system that was previously used to assess the Eigensystem Realization Algorithm (ERA) and Enhanced ERA. Eigenvalues and mode shapes are compared for each algorithm. Another example analyzes simulated data for a cantilever beam with three suspended one-degree-of-freedom subsystems, in which the parameters are adjusted to bring two natural frequencies into close proximity. The results suggest that MIA is more accurate, and more robust in the treatment of noisy data, than either ERA version, and that it is able to identify modes whose bandwidth is comparable to the difference of adjacent natural frequencies.