White lupin improvement is challenged by the need to select for low seed content of total quinolizidine alkaloids (QAs) when crossing low-alkaloid (sweet-seed) with bitter-seed (landrace) material. This study, which focused on 45 international landraces and 142 broadly sweet-seed breeding lines, aimed at (a) assessing the ability of Near-Infrared Spectroscopy (NIRS) to distinguish broadly sweet-seed from bitter-seed material and, possibly, lines with particularly low QA content within broadly sweet-seed material; and (b) comparing landrace and breeding material in terms of the composition and amount of QA compounds. QA content was analyzed using a gas chromatography–mass spectrometry method. NIRS analyses were performed either on whole-seed samples or ground samples. The range of variation for total QA was 95–990 mg/kg among breeding lines and 14,041–37,321 among landraces. NIRS was able to discriminate broadly sweet-seed from bitter-seed material when using flour samples, non-destructive 10-seed samples, and even individual whole seeds (with <1% misclassification). It was unable to identify material with particularly low QA content. Landrace and breeding line germplasm differed in the proportions of individual QAs. Patterns of geographical variation for total QA content of landraces were identified. Our results can contribute to define an efficient NIRS-based pipeline to select for low total QA content.