The resolution of magnetic resonance images reconstructed using the discrete Fourier transform (DFT) algorithm is limited by the effective window generated by the finite data length. The transient error reconstruction approach (TERA) is an alternative reconstruction method based on autoregressive moving average (ARMA) modeling techniques. Quantitative measurements comparing the truncation artifacts present during DFT and TERA image reconstruction show that the modeling method substantially reduces these artifacts on "full" (256 X 256), "truncated" (256 X 192), and "severely truncated" (256 X 128) data sets without introducing the global amplitude distortion found in other modeling techniques. Two global measures for determining the success of modeling are suggested. Problem areas for one-dimensional modeling are examined and reasons for considering two-dimensional modeling discussed. Analysis of both medical and phantom data reconstructions are presented.