We obtained multiwavelength coverage of the soft X-ray transient GRO J1655-40 during its 1996 outburst, using HST, RXTE, CGRO, and ground-based facilities. This outburst was qualitatively different from other SXT outbursts and from previous outbursts of this source. The onset of hard X-ray activity occurred very slowly, over several months, and was delayed relative to the soft X-ray rise. During this period, the optical fluxes declined steadily. This apparent anticorrelation is not consistent with with the standard disk instability model of SXT outbursts, nor is it expected if the optical output is dominated by reprocessed X-rays, as in persistent low-mass X-ray binaries. .