We investigate the amount of scatter in the Tully-Fisher relation (TFR) when using optical long-slit \ha\ rotation curves to determine the velocity widths of spiral galaxies. We study a sample of 25 galaxies in the Coma region of the sky which were shown in Bernstein et al. \markcite{B1} (1994) to exhibit an extraordinarily low scatter of 0.10 mag RMS in the I magnitude vs 21-cm width TFR. Using the same I magnitudes with new widths derived from high-quality \ha\ rotation curves, we measure an RMS scatter of 0.14 mag in the TFR. This suggests that measurement errors and ``astrophysical errors'' (such as non-circular gas motion) on the \ha\ velocity widths are below 6%, and optical widths are nearly as good for TFR studies as 21-cm widths. The scatter and form of the TFR are found to be robust under choice of velocity width-extraction algorithm, as long as the central portions of the optical rotation curve are ignored and low-S/N points are not weighted too heavily. In this small sample there is no evidence that rotation curve shapes vary systematically with rotation velocity, nor that rotation curve shape can be used to reduce the scatter in the TFR.