The Colorado River has flowed across the dextral strike-slip plate boundary between the North American and Pacific plates since the latest Miocene or earliest Pliocene. The Fish Creek-Vallecito Basin (FCVB) lies on the Pacific Plate in southern California, dextrally offset from the point where the modern Colorado river enters the Salton Trough; it contains a record of ancestral Colorado River sedimentation from 5.3–2.5 Ma. The basin stratigraphy exhibits a changing balance between locally derived (L-Suite) and Colorado River (C-Suite) sediments. This paper focuses on the Palm Springs Group (PSG), a thick fluvial and alluvial sequence deposited on the upper delta plain (between 4.2–2.5 Ma) when the Colorado was active in the area, allowing the detailed examination of the processes of sediment mixing from two distinct provenance areas. The PSG consists of three coeval formations: 1) Canebrake Conglomerate, a basin margin that has coarse alluvial fan deposits derived from surrounding igneous basement; 2) Olla Formation, fan-fringe sandstones containing L-Suite, C-Suite, and mixed units; and 3) Arroyo Diablo Formation, mineralogically mature C-Suite sandstones. Stratigraphic analysis demonstrates that the river flowed through a landscape with relief up to 2000 m. Satellite mapping and detailed logging reveal a variable balance between the two suites in the Olla Formation with an apparent upward increase in L-Suite units before abrupt cessation of Colorado sedimentation in the basin. Stable heavy mineral indices differentiate L-Suite (high rutile:zircon index: RZi 40–95) from C-Suite (RZi: 0–20). Both suites have garnet:zircon index (GZi) and apatite:tourmaline index (ATi) mostly above 50, although many L-suite and mixed Olla samples have much lower ATi (20–50), suggesting that the distal floodplain was wet and the local sediment had a longer residence time there, or went through several cycles of erosion and redeposition. Heavy mineral analysis, garnet geochemical analysis, and detrital zircon U-Pb age spectra allow us to quantify the amount of mixing from different sediment sources. These data show that about 30% of the mixed units are derived from the Colorado River and that up to 20% of the L-Suite is also derived from the Colorado River, suggesting that there was mutual cannibalisation of older deposits by fluvial channels in a transitional area at the basin margin. Although this study is local in scope, it provides an insight into the extent and nature of sediment mixing in a two-source system. We conclude that most ‘mixing’ is actually interbedding from separate sources; true mixing is facilitated by low subsidence rates and the rapid migration of fluvial channels.