We investigate the recently discovered NICMOS count-rate dependent non-linearity (Bohlin et al. 2005) using the flatfield lamps to artificially increase the count rate. A star cluster field was imaged in a lamp off-on-off sequence in all cameras in a selected set of filters, followed by a series of darks to investigate persistence and to clean the images from any remaining charge for the next orbit. Subtracting the lamp-off images from the lamp-on images clearly shows residual ADUs at the star positions, indicating that a higher background (and thus total) count rate increases the number of ADUs registered from an object. We model the non-linearity with a power law (count-rate ∝ flux) and fit this model to the data. Both NIC1 and NIC2 (NIC3 was not tested in this program) show non-linearity, becoming stronger at shorter wavelengths, but with larger amplitude than predicted by the Bohlin et al. NIC3 measurements. The non-linearity in NIC1 and NIC2 amounts to 0.06-0.10 mag offset per factor ten change in incident flux for the shortest wavelength (F090M and F110W), about 0.03 mag/dex at F160W, and less at longer wavelengths. Archival data from Cycle 7 are also analyzed, showing that the nonlinearity has not changed in NIC2 F110W, and suggesting that this effect is independent of detector temperature.