We present a review of elastography from a systems point of view. We show that elastography can be viewed as a cascade of two distinct processes. The first process involves the mapping of the distribution of local elastic moduli in the target into a distribution of local longitudinal strains. This process is governed by the theory of elasticity as applied to a particular experimental setup under some specific boundary conditions and some assumptions. Since this process involves errors due to the simplified mechanical model used, artifacts such as target hardening, stress concentrations, and limited contrast‐transfer efficiency are usually encountered. These errors may be recursively minimized by solving the inverse problem, thus increasing the contrast‐transfer efficiency such that a more accurate modulus image may be obtained. The second process involves the production of the strain image (elastogram) from ultrasonically estimated values of local strains. Here, the limitations of the ultrasound system [such as time‐bandwidth product, center frequency, and sonographic signal‐to‐noise ratio (SNR)] as well as the signal‐processing algorithms used to process the signals cause additional corruption of the data through the introduction of constraints in the attainable elastographic SNR, resolution, sensitivity, and strain dynamic range. This process is described in terms of a stochastic strain filter. These two system components are discussed in detail, and it is concluded that both must be optimized in a specific order to result in quality elastograms. © 1997 John Wiley & Sons, Inc. Int J Imaging Syst Technol, 8, 89–103, 1997