Fundamental changes in farming systems occurred throughout the medium-rainfall zone of southern Australia during the late 1990s. Pulse and canola crops replaced pastures and long-fallowing, and minimal-tillage practices were increasingly adopted. An experiment was established in 1998 to examine long-term effects of these changes on crop productivity and soil fertility. Nine rotation–tillage treatments based on 3- and 6-year cycles were compared by using wheat (Triticum aestivum) as a bioassay crop over 2001–17. Seasonal conditions during the study ranged from the Millennium Drought to the top rainfall decile. Averaged across the 17 seasons, wheat yields were significantly lower in rotations based on continuous wheat (1.73t/ha) or 3 years of perennial lucerne (Medicago sativa) (1.93t/ha) and highest after traditional long-fallow (2.92t/ha) or vetch (Vicia sativa) green manure (2.57t/ha). Wheat yields following a pulse (2.23t/ha) or canola (Brassica napus) (2.21t/ha) were intermediate. Whereas rotation effects varied with seasonal rainfall, there was a long-term trend for relative yields in continuous wheat rotations to decline and those following a green manure or fallow to differ increasingly from other treatments. Compared across the same rotation (canola–wheat–pulse), average wheat yields under no-tillage (2.09t/ha) were significant lower (P<0.05) than under reduced (2.22t/ha) and conventional (2.29t/ha) tillage. By contrast, grain protein concentration increased with increasing proportion and type of legumes (green manure and pasture>pulse) in the rotation via their effect on soil mineral nitrogen (N). Lowest protein was recorded in continuous wheat and highest where at least one-third of the rotation contained a non-pulse legume, i.e. vetch green manure, lucerne or annual medic (Medicago truncatula). Soil-borne cereal fungal pathogens and nematodes generally had little effect on grain yields. Soil organic carbon (SOC) and total N stocks declined in all treatments over time, except in the lucerne and green manure treatments, where total N was maintained. Greatest declines in SOC and total N occurred in the fallow treatment. Results from the first 20 years of experimentation suggest that differences in grain yield (and protein) of wheat were most likely the result of treatment (rotation and tillage practice) effects on soil nitrate and water (growing-season rainfall and fallow storage) rather than disease. We found no evidence that current farming systems based on use of pulses, canola and reduced tillage are less able to maintain intervening cereal yields than traditional systems. On the contrary, practices such as long-fallowing are expected to have increasing negative influence on productivity in the longer term via decreased soil C and N fertility.