The aim of this work was to test a surveillance protocol able to detect extended-spectrum β-lactamase (ESBL)-, cephalosporinase (AmpC)- and carbapenemase (CP)-producing gram-negative bacteria in three conveniently chosen dairy farms with known prior occurrences of ESBL- and CP-producing strains. The protocol was applied monthly for a year. At each visit, 10 healthy lactating dairy cows were rectally swabbed, and raw milk filters (RMFs) were sampled in two of the three farms. Bacterial isolation was based on a first screening step with MacConkey agar supplemented with 1 mg/L cefotaxime and commercial carbapenem-supplemented media. We failed to detect CP-producing strains but showed that ESBL-Escherichia strains, found in one farm only (13 strains), were closely associated with multi-drug resistance (12 out of 13). The limited number of conveniently selected farms and the fact that RMFs could not be retrieved from one of them limit the validity of our findings. Still, our results illustrate that ESBL-status changes monthly based on fecal swabs and negative herds should be qualified as “unsuspected” as proposed by previous authors. Although surveillance of farm statuses based on RMF analysis could theoretically allow for a better sensitivity than individual swabs, we failed to illustrate it as both farms where RMFs could be retrieved were constantly negative. Determination of CP herd-level status based on RMFs and our surveillance protocol was hindered by the presence of intrinsically resistant bacteria or strains cumulating multiple non-CP resistance mechanisms which means our protocol is not specific enough for routine monitoring of CP in dairy farms.