We have searched 4.5 square degrees of archival HST/ACS images for cosmic strings, identifying close pairs of similar, faint galaxies and selecting groups whose alignment is consistent with gravitational lensing by a long, straight string. We find no evidence for cosmic strings in five large-area HST treasury surveys (covering a total of 2.22 square degrees), or in any of 346 multi-filter guest observer images (1.18 square degrees). Assuming that simulations ccurately predict the number of cosmic strings in the universe, this non-detection allows us to place upper limits on the unitless Universal cosmic string tension of G mu/c^2 < 2.3 x 10^-6, and cosmic string density of Omega_s < 2.1 x 10^-5 at the 95% confidence level (marginalising over the other parameter in each case). We find four dubious cosmic string candidates in 318 single filter guest observer images (1.08 square degrees), which we are unable to conclusively eliminate with existing data. The confirmation of any one of these candidates as cosmic strings would imply G mu/c^2 ~ 10^-6 and Omega_s ~ 10^-5. However, we estimate that there is at least a 92% chance that these string candidates are random alignments of galaxies. If we assume that these candidates are indeed false detections, our final limits on G mu/c^2 and Omega_s fall to 6.5 x 10^-7 and 7.3 x 10^-6. Due to the extensive sky coverage of the HST/ACS image archive, the above limits are universal. They are quite sensitive to the number of fields being searched, and could be further reduced by more than a factor of two using forthcoming HST data.