We describe the internal photometric calibration of the Deep Lens Survey, which consists of five widely separated fields observed by two different observatories. Adopting the global linear least-squares (‘ubercal’) approach developed for the Sloan Digital Sky Survey (SDSS), we derive flat-field corrections for all observing runs, which indicate that the original sky flats were non-uniform by up to 0.13 mag peak-to-valley in z band, and by up to half of that amount in BVR. We show that the application of these corrections reduces spatial non-uniformities in corrected exposures to the 0.01–0.02 mag level. We conclude with some lessons learned in applying ubercal to a survey structured very differently from SDSS, with isolated fields, multiple observatories and shift-and-stare rather than drift-scan imaging. Although the size of the error caused by using sky or dome flats is instrument and wavelength dependent, users of wide-field cameras should not assume that it is small. Pipeline developers should facilitate routine application of this procedure, and surveys should include it in their plans from the outset.