本文嘗試從“直接途徑”與“間接途徑”的對比去探討漢字（以形聲為主）的認知特性。文中首先擬設一個“內部詞庫”的結構模型，在此模型中漢字形、音、義三項要素處於半獨之的連繫狀態。以此模型去說明由“目睹字形”到提取詞義的直接提取和間接提取兩種途徑。其中間接提取又再根據字形──語音轉換過程的不同而再細分為分析性、綜合性及半分析半綜合三種方式。 由於國人偏重於字形訊息，所以在解讀漢字中字形訊息有較大的提示效度，國人亦因此更多採用直接途徑。 本文亦對解讀漢字時出現語音中介這種說法提出質疑，指出所謂語音中介其實只不過是語音激發而已。

　　　　　The article attempts to clarify the cognitive features of Chinese graphs (mainly xingsheng 形聲) form the perspective of direct access vs indirect access. A paradigm of internal lexicon is set up. In the paradigm, "graphic", "phonetic" and "semantic" components of a Chinese graph are connected to each other in an semi-independent fashion. The paradigm is used to account for the direct access and indirect access form perceptual stimulus to lexical meaning. The indirect access is further classified into three according to the graphemic--phonemic process, i.e., analytic, synthetic and semi-analytic & semi-synthetic. As Chinese readers tend to rely more on graphic information, graphic information enjoys greater cue validity; which explains why direct access plays a greater role in decoding of Chinese graphs. The article also queries the thesis that there is phonological mediation in decoding of chinese graphs. It demonstrates that the so-called phonological mediation is no more than phonological activation.