Traditional　object　classification　in　3D　point　cloud　scenes　relies　heavily　on　large-scale　labeled　training　data,　which　is　both　time-consuming　and　labor-intensive　to　obtain.　Unsupervised　Domain　Transfer　(UDT)　mitigates　this　challenge　by　transferring　knowledge　from　a　labeled　source　domain　to　an　unlabeled　target　domain.　However,　existing　UDT-based　methods　often　require　complex　neural　architectures　and　substantial　computational　resources.　This　letter　proposes　a　novel　UDT　framework　that　integrates　hierarchical　prompt　learning　with　a　3D　foundational　model.　The　proposed　method　consists　of　a　modality　alignment　stage　and　an　unsupervised　transfer　stage.　In　the　modality　alignment　stage,　cross-modal　hierarchical　prompts　are　employed　to　align　the　Visual-Language　(V-L)　modality　in　the　3D　foundational　model　through　a　V-L　coupling　module.　In　the　unsupervised　transfer　stage,　cross-domain　hierarchical　prompts　and　a　Target-Source　(T-S)　coupling　module　facilitate　the　alignment　of　multi-scale　contextual　information　across　domains,　ensuring　efficient　and　accurate　knowledge　transfer.　Extensive　experiments　conducted　on　multiple　datasets　collected　from　various　laser　scanners　demonstrate　the　effectiveness　of　our　proposed　approach.　©　1994-2012　IEEE.