There is a need for continued assessment of methane (CH4) emissions associated with natural gas (NG) production, especially as recent advancements in horizontal drilling combined with staged hydraulic fracturing technologies have dramatically increased NG production (we refer to these wells as "unconventional" NG wells). In this study, we measured facility-level CH4 emissions rates from the NG production sector in the Marcellus region, and compared CH4 emissions between unconventional NG (UNG) well pad sites and the relatively smaller and older "conventional" NG (CvNG) sites that consist of wells drilled vertically into permeable geologic formations. A top-down tracer-flux CH4 measurement approach utilizing mobile downwind intercepts of CH4, ethane, and tracer (nitrous oxide and acetylene) plumes was performed at 18 CvNG sites (19 individual wells) and 17 UNG sites (88 individual wells). The 17 UNG sites included four sites undergoing completion flowback (FB). The mean facility-level CH4 emission rate among UNG well pad sites in routine production (18.8 kg/h (95% confidence interval (CI) on the mean of 12.0-26.8 kg/h)) was 23 times greater than the mean CH4 emissions from CvNG sites. These differences were attributed, in part, to the large size (based on number of wells and ancillary NG production equipment) and the significantly higher production rate of UNG sites. However, CvNG sites generally had much higher production-normalized CH4 emission rates (median: 11%; range: 0.35-91%) compared to UNG sites (median: 0.13%, range: 0.01-1.2%), likely resulting from a greater prevalence of avoidable process operating conditions (e.g., unresolved equipment maintenance issues). At the regional scale, we estimate that total annual CH4 emissions from 88 500 combined CvNG well pads in Pennsylvania and West Virginia (660 Gg (95% CI: 500 to 800 Gg)) exceeded that from 3390 UNG well pads by 170 Gg, reflecting the large number of CvNG wells and the comparably large fraction of CH4 lost per unit production. The new emissions data suggest that the recently instituted Pennsylvania CH4 emissions inventory substantially underestimates measured facility-level CH4 emissions by >10-40 times for five UNG sites in this study.