We reinvestigate the massive stellar content of the irregular dwarf galaxy NGC 3125 (Tol 3) using Very Large Telescope (VLT)/FORS1 imaging and spectroscopy, plus archival VLT/ISAAC, Hubble Space Telescope (HST)/FOC and HST/STIS data sets. FORS1 narrow-band imaging confirms that the NGC 3125-A and -B knots represent the primary sites of Wolf–Rayet (WR) stars, whilst HST imaging reveals that both regions host two clusters. Both clusters within region A host WR stars (A1 and A2), for which the optically fainter cluster A2 is heavily reddened. It is not clear which cluster within region B hosts WR stars. Nebular properties are in good agreement with previous studies and infer a Large Magellanic Cloud (LMC) like metallicity of log(O/H)+12∼8.3. LMC template mid-type WN and early-type WC spectra are matched to the observed blue and red WR bumps of A1 and B, permitting the contribution of WC stars to the blue bump to be quantified. From our FORS1 spectroscopy, we obtain N(WN5–6:WC4) = 105:20, ∼55:0 and 40:20 for the clusters A1, A2 and B1 + 2, respectively. Our results are a factor of ∼3 lower than previously reported by optical studies as a result of a lower Hα/Hβ derived interstellar reddening. Using Starburst99 theoretical energy distributions to estimate O star populations for each cluster, we find N(WR)/N(O) = 0.2 for A1 and 0.1 for A2 and the clusters within region B. From Hα narrow-band imaging, the O star content of the Giant H II regions A and B is found to be a factor of 5–10 times higher than that derived spectroscopically for the ultraviolet (UV)/optically bright clusters, suggesting that NGC 3125 hosts optically obscured young massive clusters, further supported by VLT/ISAAC K-band imaging. Archival HST/STIS UV spectroscopy confirms the low interstellar reddening towards A1, for which we have determined a Small Magellanic Cloud (SMC) extinction law for NGC 3125, in preference to an LMC or starburst law. We obtain N(WN5–6) = 110 from the slit loss corrected He II λ1640 line flux. This is in excellent agreement with optical results, although it is a factor of 35 times lower than that inferred from the same data set by Chandar, Leitherer & Tremonti. The discrepancy is due to an anomalously high interstellar reddening derived from their use of the generic starburst extinction law. Highly discrepant stellar populations may result in spatially resolved star-forming regions from UV and optical studies through the use of different extinction laws.