本研究利用RAPD分子標誌探討由臺南區農業改良場義竹工作站所提供之大蒜種原50種(分別來自臺灣、大陸、埃及、印度、印尼及菲律賓);在農試所試驗田區採集之15個臺灣地方品種及二個韓國品種的嫩葉樣品;以及臺大農藝系所收集之11個大陸品種等，一共80個大蒜品種(系)間的遺傳相關性。同時在臺大園藝分場觀察各種原的外表性狀。多數品種為葉姿較直立的硬骨蒜，假莖白色、葉色綠或濃綠、白色鱗莖但可能蒜瓣外皮帶紫色，這些質的性狀並不足以區分各品種。以RAPD的17條引子擴增出清晰、容易判讀且具穩定再現性之標誌共59個，其中23個為對照材料特有，僅36個條帶在大蒜品種(系)間具多型性(polymorphic)，平均每引子可擴增出2.1組多型性標誌。80個大蒜供試品種(系)的遺傳相似度(genetic similarity)介於0.16-1.00。由群叢分析結果將參試品種(系)在平均遺傳相似度0.26分為三大群，第一群群內遺傳相似性高，大部分臺灣品種(系)在此群且無法區分，包括有蒜頭用‘大片黑’類及‘和美蒜’;‘新品種黑葉’、‘鳳山選二號’及花蒜種類與宜蘭地區品種(系)皆與‘北蒜’、中國大陸南方蒐集品種(系)在另一群中;大陸北方山東、河南地區品種(系)則與菲律賓、印度等外國品種(系)同在第三大群中。相同地名的品種不一定遺傳相似性高，蒜頭用品種間遺傳相似度高，花蒜類品種間相似性較低，群叢分析結果大致與材料來源及類型或採收目的有相關性。主成分分析獲得之前三個主成分可解釋84%的變異量，品種(系)在前兩個主成分解釋79.6%變異的分佈與群叢分析結果類似。

RAPD makers were used to analyze the genetic variations of 80 garlic accessions mostly from Taiwan, including 50 collected by Yi-Chu Branch, Tainan DAIS, 17 from TARI and 11 from China. Fifty germplasm accessions with bulbs were planted at horticultural experiment farm of NTU to observe the characteristics of their plants and bulbs. Most germplasms were of bulb type with somewhat erect leaves, pseudo-stem in white color, green to dark green foliage, white bulb scales with some clove outer scale in purple tinge. The morphological differences were not sufficient to distinguish garlic line. The constructed dendrogram divided all 80 garlic accessions into three groups at the genetic similarity of 0.26. Most Taiwan lines including bulb-use varieties were clustered into the first group with very high genetic similarity. Flower type (scape-use), ‘FengShan Sel No. 2’ and Yi-Lan landraces were grouped with varieties from Hu-Nan and Yun-Bab if China in the second cluster. Varieties from northern China being different from Taiwan varieties were clustered together with other foreign varieties of India and the Philippines. Among Taiwan germplasms and lines, those of the same region (have a common name) may not share high similarities. In general bulb use varieties have small genetic difference while flowering types have more genetic variation. In the principle component analysis, the first two components could explain 79.6% of variability and the two-dimension distribution depicts the same results from cluster analysis.