The efficacy of the GHOST proximity correction technique for achieving submicron linewidth control in electron beam lithography has been experimentally evaluated. Using the AT&T electron beam exposure systems, the GHOST parameters (correction dose, defocused beam diameter, and overlay accuracy) required for proximity exposure correction at 20 kV were determined via scanning electron microscopy and electrical linewidth measurement techniques. Optimum proximity correction conditions were established from a matrix of correction doses and correction beam diameters on two different substrates: chromium on glass and aluminum on silicon. On chromium, linewidth control of better than ±0.05 μm was obtained using a 3‐μm beam and a correction dose between 30% and 40% of incident, or with a 4‐μm beam and a 40% correction dose. Even better proximity correction was achieved on aluminum with a 5‐μm defocused beam and a 30% correction dose. In conjunction with the experimental data, a computer model was developed to interpret these results. The model assumes an elliptical profile for the incident GHOST beam rather than a Gaussian. Using this model, the tolerances on correction dose and defocused beam diameter for 0.5‐μm lithography on aluminum were determined to be 9% and 1.6 μm, respectively. The third GHOST parameter investigated, overlay accuracy between the incident and GHOST exposures, was also found to be critical. A 0.5‐μm overlay error resulted in a 0.05‐μm linewidth deviation across a 0.5‐μm line and space pattern. Last, it was found that with AZ‐2400‐17 resist the loss of resist thickness and developer latitude upon GHOSTing will not result in pinholes or degrade process latitude as long as the correction dose is maintained at ≤ 40% of the incident dose, and a dilute developer is employed.