We use ~88 arcmin2 of deep (26.5 mag at 5σ) NICMOS data over the two GOODS fields and the HDF-South to conduct a search for bright z 7 galaxy candidates. This search takes advantage of an efficient preselection over 58 arcmin2 of NICMOS H 160-band data where only plausible z 7 candidates are followed up with NICMOS J 110-band observations. ~248 arcmin2 of deep ground-based near-infrared data (25.5 mag, 5σ) are also considered in the search. In total, we report 15 z 850-dropout candidates over this area—7 of which are new to these search fields. Two possible z ~ 9 J 110-dropout candidates are also found, but seem unlikely to correspond to z ~ 9 galaxies (given the estimated contamination levels). The present z ~ 9 search is used to set upper limits on the prevalence of such sources. Rigorous testing is undertaken to establish the level of contamination of our selections by photometric scatter, low-mass stars, supernovae, and spurious sources. The estimated contamination rate of our z ~ 7 selection is ~24%. Through careful simulations, the effective volume available to our z 7 selections is estimated and used to establish constraints on the volume density of luminous (L* z = 3, or ~–21 mag) galaxies from these searches. We find that the volume density of luminous star-forming galaxies at z ~ 7 is 13+8 –5 times lower than at z ~ 4 and >25 times lower (1σ) at z ~ 9 than at z ~ 4. This is the most stringent constraint yet available on the volume density of L* z = 3 galaxies at z ~ 9. The present wide-area, multi-field search limits cosmic variance to 20%. The evolution we find at the bright end of the UV LF is similar to that found from recent Subaru Suprime-Cam, HAWK-I or ERS WFC3/IR searches. The present paper also includes a complete summary of our final z ~ 7 z 850-dropout sample (18 candidates) identified from all NICMOS observations to date (over the two GOODS fields, the HUDF, galaxy clusters).