Advancements in integrated circuit (IC) package technology are increasingly leading to size shrinkage of modern microelectronic packages. This size reduction presents a challenge for the detection and location of the internal features/defects in the packages, which have approached the resolution limit of conventional acoustic microimaging, an important nondestructive inspection technique in the semiconductor industry. In this paper, to meet the challenge the learning overcomplete representation technique is pursued to decompose an ultrasonic A-scan signal into overcomplete representations over a learned overcomplete dictionary. Ultrasonic echo separation and reflectivity function estimation are then performed by exploiting the sparse representability of ultrasonic pulses. An improved acoustic microimaging technique is proposed by integrating these operations into the conventional acoustic microimaging technique. Its performance is quantitatively evaluated by elaborated experiments on ultrasonic A-scan signals using acoustic microimaging (AMI) error criteria. Results obtained both from simulated and measured A-scans are presented to demonstrate the superior axial resolution and robustness of the proposed technique.