 Pronoun resolution has been an important issue in reading research.  Many studies found the phenomena of first mention advantage in pronoun resolution, but Järvikivi, et al (2005) challenged the first mention advantage by questioning the percentage of first visits critical characters and the number of fixations on critical characters at different time sets.  These different conclusions may results from using different statistical techniques, e.g., ANOVA analysis versus mixed- effects modeling.  Mixed-effects modeling treat subjects and items in an experiment as random samples from larger populations (Raaijmakerm, et al.,1999). Moreover, because the residuals of items within the same subject are likely to be correlated, it is inappropriate to use the typical regression analysis (Baayen, et al., 2008).  Mixed models should be adopted for analyzing data from repeated measures experiments (Bagiella, Sloan, & Heitjan, 2000). In our opinion, the mixed-effects analysis has not yet been widely applied in the research of language.  The purpose of this study is to compare different analysis methods in eye tracking data. The first part examined the fixation time as an outcome variable through the use of random coefficients models. The second part analyzed fixation counts by using the GEE and GLMM approaches, respectively. Methods Analysis of eye movements data Using Mixed Effects Modeling and Poisson Regression Chiu Hua Huang, 1 Wei Ming Luh, 2 Ching-Fan Sheu 3, Yuhtsuen Tzeng, 4 Minglei Chen 5 1, 2, 3 Institute of Education, National Cheng Kung University, Taiwan 4 Institute of Curriculum Studies, Nation Chung Cheng University, Taiwan 5 Institute of Learning and Instruction, National Central University, Taiwan Participants and apparatus  We recruited 32 college students at National Chung Cheng University in southern Taiwan.  The experiment lasts about 30 minutes. An SR Research EYELINK Experimental Builder version eyetracker was used. Materials and design  The study used 2×2 design of order-of-mention (first- and second-mentioned) and gender cue (male or female).  Thirty-six short texts were constructed and each contained two sentences and an interrogative question. To describe the event, the first sentence was designed to begin with two names do certain event. The second sentence started with a pronoun with a gender cue, subsequent to the event described in the first sentence. The third sentence was an interrogative question to test if readers could determine the pronoun by gender cue provided in sentence two.  Additional six practice trials and 36 fillers. Procedures  All sentences were presented in the middle of the computer by the format of the self-paced reading task (Yang et al., 1999). On each trial, after the participants had fixated on the fixation point, participants needed to press space bar to begin the first screen of the trial. Introduction Results Four experimental conditions: First-mention refer to Man Sam is in the hotel lobby and Jane buys the tickets. He goes to Taipei. Second-mention refer to Man Jane is in the hotel lobby and Sam buys the tickets. He go to Taipei. First-mention refer to Female Jane is in the hotel lobby and Sam buys the tickets. She go to Taipei. First-mention refer to Female Sam in-the-hotel-lobby and Jane buys the tickets. She go to Taipei.  Random coefficients models  The predictor variable included design factors and first gaze duration of sentences 2 (S2FGD)  The outcome variable is second gaze duration of S1 (S1SG), and the predictor variables are gender cue (GC), order-of-mention(Order), and S2FGD.  The outcome variable (S1SG) was not normally distributed (Shapiro- Wilk Statistic =.802, p=.000), therefore it was log-transformed. Finally, we fitted the mixed effects model to data by using SAS 9.2. The output is show in Table 1 and Table 2.  It is shown that when we considered subjects and items as random effects, the order effect disappeared, but the effect of S2FGD remained. Therefore, the first mention advantage has not been confirmed in the above analysis.  Poisson models and GLMM approach  The outcome variable here was the second gaze count of sentence 1 (S1SGC). Predictor variables include gender cue (GC), order-of- mention (Order), First gaze count of sentence 2 (S2FGFC). The distribution of S1SGC was positively skewed.  Poisson models  We examined two approaches: The generalized estimating equations (GEE) and the generalized linear mixed models (GLMM). The result is presented in Table 3, showed none of the three predictors were statistically significant.  GLMM approach  The output of Fit Statistics model shows in Table 4 and 5, respectively.  GLMM results showed that no significant effects was found for any predictors. Conclusion There are two main findings in the current study. First, by using mixed model for reading time, we found that there is no significance in the order effect, but the first gaze duration of sentence 2 (S2FGD) significantly predicted the effect. Second, when we examined the fixation counts while readers rereading the first sentence, we did not find any order effect from the analysis of GEE or GLMM. The results from new analysis approach are clearly incompatible with the first mention advantage, which previous researches of pronoun resolution have claimed.