The purpose of this study was to investigate online processes of assessing writing through rater’s the eye-movements. Rater’s eyes-movements are measured during rating essays, and then their fixation counts and fixation duration between a rubric and essays are analysed that change as a time-series. Our data suggests that raters showed about 58% of fixation counts and 28% of fixation duration in rubric, and they fixed during whole rating processes 41.4 seconds at essay(AOI 2) and 16.2 seconds at rubric(AOI 1); fixation counts have been in sharp decline until they read the 10th essay, and then changed gentle decline trends following. We made a regression model of the log function base on relative fixation count and fixation duration data. We estimated a 88th essay as the predicted value that fixation rate is a zero. This means a time point that raters construct a cognitive representation of the rubric. And also we discussed that practices should require about 25 minimum rating sample essay for rater training that start long-tail trend line on our non-linear regression. This means a practice effect is slight that is represent the inclination declines on the graph.