This paper evaluates deconvolution (inverse filtering) as applied to ultrasonic imaging systems, and discusses the obstacles which are encountered employing the technique in practice. A minicomputer is used to generate artificial echo signals, simulating rf signals resulting from a set of point reflectors in a homogeneous medium, as recorded by an electronically focused group-steered linear array scanner. Two-dimensional deconvolution in combination with a Wiener noise reduction filter (i.e., a Wiener-Inverse filter) is applied to these simulated rf signals, which were contaminated with white noise. The efficacy of the Wiener-Inverse filter is defined in terms of its ability to resolve two point reflectors with a lateral spacing equal to the local -6 dB width of the ultrasonic beam. In favorable circumstances, the targets are resolved at signal-to-noise ratios (SNR) better than 20 dB, where SNR is defined as the maximum signal power divided by the average noise power level. Nonlinear effects due to quantization or signal clipping are investigated. In order to improve the resolution of an rf signal with a dynamic range of 40 dB, the input signal should be digitized at a minimum of 12 bits. The problem of signal clipping can be circumvented by oversampling. The two-dimensional Wiener-Inverse filter is defined in terms of both temporal and spatial properties of the insonification. Effects of wave diffraction give rise to a depth-dependent ultrasonic beam. As a result of a misfit of the Wiener-Inverse filter and the local properties of the ultrasonic beam, erroneous noisy texture arises in the image. Adaptation of the Wiener-Inverse filter with respect to the beam properties gives acceptable results, at the expense of a rather large computational effort.