Aberrations in ultrasound imaging are modeled as a linear filter bank. The response of each linear filter is determined through blind system identification and is used to correct the aberration by applying a predistortion to the transmit signals and a compensation to the receive signals. To reduce computation, the blind system identification can be done through multiple levels. In a two-level implementation, the transducers in the array are grouped into subapertures. The blind system identification is applied in each of the subapertures to obtain the first-level response to each element. Using these responses, a common input is found for each subaperture, and blind system identification is applied to find the second-level responses. From the first-level and second-level responses, the response at each filter in the bank is determined.