This dataset details grain size and bulk mineralogy from marine sediment core MD07-3119, collected in the Southeastern Pacific (-46.083167°N, -76.099833°E) at 2523 m water depth during cruise MD159 (IMAGES XV - Pachiderme). The combined dataset spans depth intervals from 541 cm to 3245 cm, corresponding to 16 to 42 ka cal BP (calculated using the age-depth model from Andrzejewski et al. (2026), submitted, and Lebeaupin et al. (in prep.), submitted). A total of 161 samples (every 10-20 cm) were analysed.Grain-size data were obtained using a Malvern Mastersizer 3000 laser-diffraction particle-size analyser equipped with a Hydro LV dispersion unit. To isolate terrigenous material, samples underwent sequential chemical treatments to remove organic matter using 35% hydrogen peroxide (10 mL), carbonates with 10% hydrochloric acid (5 mL), and biogenic silica with 2 N sodium hydroxide (5 mL). After these treatments, 2% sodium hexametaphosphate (5 mL) was added to achieve complete particle disaggregation. During analysis, a stirrer set to 1000 rpm was used to maintain sample suspension, and continuous ultrasounds were applied. Grain-size distributions were determined from three consecutive 20-second runs, and modal values were calculated using Mastersizer v3.88. The mean grain size was calculated using the Gradistat v9.1 software (Blott and Pye, 2001). The mean grain size was calculated on the <150 um fraction only (Mean w/o IRD). Semi-quantitative relative abundances (%) of Quartz, Plagioclase, Illite, Kaolinite, Calcite, K-feldspar, Amphibole, Olivine, and Pyroxene were analysed using a Bruker D8 Advance X-ray Diffractometer (XRD). Powdered samples were prepared by grinding to a fine, homogeneous size to ensure accurate phase identification and quantification. The powders were mounted as unoriented mounts to minimise diffraction bias and subsequently scanned by XRD over the 2θ range of 5° to 70°. Mineral identification and semi-quantitative analysis were performed using the DIFFRAC.EVA software. Semi-quantitative abundances were estimated (±5 wt. %) using the Reference Intensity Ratio (RIR) method, following Cook et al. (1975).