Relative sea-level changes during the last ∼2500 years in New Jersey, USA were reconstructed to test if late Holocene sea level was stable or included persistent and distinctive phases of variability. Foraminifera and bulk-sediment δ¹³C values were combined to reconstruct paleomarsh elevation with decimeter precision from sequences of salt-marsh sediment at two sites using a multi-proxy approach. The additional paleoenvironmental information provided by bulk-sediment δ¹³C values reduced vertical uncertainty in the sea-level reconstruction by about one third of that estimated from foraminifera alone using a transfer function. The history of sediment deposition was constrained by a composite chronology. An age–depth model developed for each core enabled reconstruction of sea level with multi-decadal resolution. Following correction for land-level change (1.4 mm/yr), four successive and sustained (multi-centennial) sea-level trends were objectively identified and quantified (95% confidence interval) using error-in-variables change point analysis to account for age and sea-level uncertainties. From at least 500 BC to 250 AD, sea-level fell at 0.11 mm/yr. The second period saw sea-level rise at 0.62 mm/yr from 250 AD to 733 AD. Between 733 AD and 1850 AD, sea level fell at 0.12 mm/yr. The reconstructed rate of sea-level rise since ∼1850 AD was 3.1 mm/yr and represents the most rapid period of change for at least 2500 years. This trend began between 1830 AD and 1873 AD. Since this change point, reconstructed sea-level rise is in agreement with regional tide-gauge records and exceeds the global average estimate for the 20th century. These positive and negative departures from background rates demonstrate that the late Holocene sea level was not stable in New Jersey.