Koegel Fontein, about 350 km north of Cape Town, is the only known early Cretaceous anorogenic igneous complex along the volcanic rifted margin of South Africa. The oldest rocks of the complex are minor granite and syenite intrusions at 144 Ma, which were followed by tholeiitic and alkaline basalt dykes, then by microsyenite and quartz porphyry dykes. The youngest and largest igneous unit is the 135 Ma Rietpoort Granite, with an exposed diameter of about 20 km. The country rocks are Mesoproterozoic gneisses of the Namaqua–Natal Province, which in many places were deformed and retrogressed by Pan-African tectonism. Whole-rock δ18O values from the Rietpoort Granite and smaller plutonic units (syenite, granite) are in the range 6–9‰ (outliers 4‰ and 17‰). Quartz δ18O values from all units are in a narrow range and indicate magma δ18O values between 6 and 8‰. In contrast to the syenites and granites, most mafic and silicic dyke units have δ18O <6‰, as low as –4·1‰. Quartz porphyry dykes that are compositionally similar to the Rietpoort Granite have a bimodal distribution of δ18O values in both whole-rock and quartz phenocrysts. The magma δ18O values estimated from the phenocryst data define a ‘normal group’ identical to the Rietpoort Granite (6–8‰) and a ‘low-δ18O group’ (0 to –2‰). The microsyenite and mafic dykes also yield negative δ18O values, but the strong hydrothermal alteration of these rocks and lack of fresh phenocrysts make a primary origin of the low δ18O values unlikely and untestable. Whole-rock δD values of igneous units and basement rocks average –99‰, which corresponds to a palaeo-meteoric water with δ18O as low as –9‰. This is much lower than the expected value for meteoric water at the time of emplacement, given the low latitude (30–40°S). Quartz veins cutting the mafic dykes have δ18O values as low as –2‰, which attest to hydrothermal fluids having low δ18O values. Country rocks in the study area have a large range of δ18O (–3 to 10‰), with the majority below the mantle value of 6‰. The low δ18O values of the country rocks, although prevalent in the roof pendant of the Rietpoort Granite, do not appear to have originated from a meteoric–hydrothermal system established by the intrusions. We suggest instead that the Koegel Fontein complex was emplaced into a structurally controlled zone in the Namaqualand basement whose δ18O values had been lowered by interaction with meteoric fluid during reactivation along Pan-African shear zones. Initial emplacement of the magmas caused thermal dehydration of the country rocks and expulsion of low-δ18O fluids. This was followed by local partial melting of the altered crust with formation of low-δ18O crustal magmas. The O isotope data provide new constraints on the crustal vs mantle source of the Koegel Fontein magmas. The Rietpoort Granite and ‘normal δ18O’ quartz-porphyry dykes crystallized from magmas with δ18O values of 7–8‰, εNd of –5 to –7, and initial 87Sr/86Sr of 0·716–0·732, which fit a model for 30–50% meta-igneous crust similar to the local Namaqua gneisses, with a minor component of low-δ18O crust. The ‘low-δ18O’ quartz porphyry magma had an identical Nd isotope composition, but lower initial 87Sr/86Sr (0·709–0·725) and δ18O (0 to –2‰), which we attribute to melting or assimilation of hydrothermally altered basement rocks with Rb and 18O depletion.