本研究針對母語為粵語的華語學習進行華語聲母及韻母的語音偏誤分析，並且探究產生偏誤的原因以及檢視Flege（1995）提出的語言學習模組（Speech Learning Model）。依據Flege的語言學習模組理，母語與外語語音系統中，相同音直接正遷移到外語中；陌生音則在外語中建立新的語音系統，故容易學習；似音則較困難學習。而對於母語為粵語的華語學習者而言，華語語音系統有幾個讀音不存在粵語語音系統裡，例如：齦顎塞擦音[tɕ]、翹舌塞擦音[tʂ]和翹舌塞擦音的送氣音[tʂʰ]，然而學習者卻產生偏誤，其偏誤主要為翹舌音和非翹舌音的混淆。另外，[n]、[l]皆存在粵語中，然學習者常常將華語中[l]讀成[n]；[ən]和[əŋ]存在於粵語中，然學習者卻混淆。本研究結果可以看出學習者的偏誤情形大致上符合Flege語言學習模組的預測。偏誤產生之原因，除了受母語遷移的影響居多，還需考量漢字對於粵語為母語者發音上的影響，以及語內語音的混淆。本文最後提出從聽力感知與發音訓練雙管齊下，使學習者發音更加精準。

This paper examined Flege's Speech Learning Model (1995) by investigating the pronunciation produced by Cantonese-speaking Chinese learners from Macau. According to Flege's model, new sounds are easy to learn; similar, but not identical sounds are difficult for second language learners. Most of the findings supported Flege's mode, while new sounds, such as [tɕ], [tʂ], [tʂʰ] are not easy for learners. The retroflex and its counter parts are confusing to the learners. Although [n] and [l] both exist in Cantonese, learners tended to pronounce [l] as [n]. In addition, [ən] and [əŋ] occur in Cantonese as well, learners are confused by these two sounds. Most of the findings in this study support Flege's model. The pronunciation errors that leaners made were mainly affected by their native language, Cantonese, as well as due to Orthographic influence. That is, the same Chinese characters in both languages are pronounced slightly differently. We suggested that language training on listening and pronunciation simultaneously should be able to improve learners' pronunciation more accurately.