We present a method for reconstructing magnetic resonance (MR) images from data acquired using echo-planar imaging (EPI) techniques. All data were acquired from a commercial scanner, the 1.5 Tesla Picker Vista HPQ MR imaging system, equipped with a special, high-performance, gradient system. Blipped echo planar imaging (BEPI) was performed, with and without digitizer pausing during data acquisition. Both sinusoidal and trapezoidal readout gradients were programmed and tested. Samples obtained from the gradient amplifier current monitors were used to calculate the approximate position of every sample obtained in the spatial-frequency (k-space) plane. A convolution function with compact support int he k-space and good rolloff in the image domain was used to resample the data onto a lattice permitting the use of fast transformation methods to the image domain. Strong ghosting was observed in the resulting images due probably to static magnetic field variation, gradient asymmetry and echo asymmetry between data lines. A piecewise-linear function was used to model the introduced ghost and hence to remove pixels in the reconstructed image which were determined to be spurious. Initial results are promising.