There are few records of long-term trends in soil C and N in grazed pasture systems but recent measurements have demonstrated unexplained losses on New Zealand lowlands. To determine whether losses were also occurring in hill country pastures, we analyzed archived soil samples collected between 1983 and 2006 from two slope classes (steep and easy) at the Whatawhata Research Centre. Soils were Ultic Hapludand and Typic Haplohumult on the easy slopes (10–20°), and Typic Haplohumult on the steeper slopes (30–40°). Soil samples (0–75 mm) had been collected from paddocks that were fertilized with six different loading rates of P (ranging from 0 to 100 kg P ha⁻¹ year⁻¹ since 1985). This range of P loadings allowed us to determine whether P inputs would regulate trends in soil C and N. While there were significant temporal trends in C and N (P < 0.05), these were not unidirectional and trends were not dependent on P loading rate. On average, soil C initially increased during the first 6 years of the trial at 0.270% C year⁻¹ (1.56 t ha⁻¹ year⁻¹) and 0.156% C year⁻¹ (1.06 t ha⁻¹ year⁻¹) on easy and steep slopes, respectively. Subsequently, there was no significant trend in soil C on the easy slopes but soil C declined at −0.066% year⁻¹ (0.45 t ha⁻¹ year⁻¹) on the steep slopes. Similarly, soil N increased between 1983 and 1989 at 0.025% N year⁻¹ (144 kg ha⁻¹ year⁻¹) and 0.012% N year⁻¹ (82 kg ha⁻¹ year⁻¹) on easy and steep slopes, respectively. Post-1989, small but significant losses of total N were measured on the steep slopes of 0.004% year⁻¹ (27 kg N ha⁻¹ year⁻¹) (P < 0.05) with no trend on the easy slopes. Two potential causal factors for these decadal-scale patterns were identified, operating via changes in primary productivity. These were lower S inputs from 1989 due to a change in fertilizer type, and a series of relatively dry summers during the 1990s. These significant inter-annual trends in soil C and N complicate attempts to measure long-term changes in soil organic matter associated with land use change and management practices. This study has demonstrated the potential error associated with infrequent soil sampling to determine long-term trends in soil C and N; large gains or losses could have been detected at Whatawhata depending on when sampling started and finished. Understanding these long-term trends in soil organic matter dynamics and driving factors requires more long-term sampling trials.