Objective To explore the characteristics and mechanisms of working memory impairment in patients with frontal lobe epilepsy (FLE) through a memory game paradigm combined with eye tracking technology.Method We included 44 patients with FLE and 50 healthy controls (HC). All participants completed a series of neuropsychological scale assessments and a short-term memory game on an automated computer-based memory evaluation platform with an eye tracker.Results Memory scale scores of FLE patients including digit span (U = 747.50, p = 0.007), visual recognition (U = 766.50, p = 0.010), and logical memory (U = 544.00, p < 0.001) were significantly lower than HC. The patients with FLE took longer to complete the four levels of difficulty of the short-term memory game than healthy controls (level 1: U = 2974.50, p = 0.000; level 2: U = 3060.50, p = 0.000; level 3: U = 2465.00, p = 0.000; level 4: U = 2199.00, p = 0.000). During the memory decoding period, first fixation on the targets took significantly longer for FLE patients for all difficulty levels compared to controls (level 1: U = 3407.00, p = 0.008; level 2: U = 3618.00, p = 0.036; level 3: U = 3345.00, p = 0.006; level 4: U = 2781.00, p = 0.000). The average fixation duration per target among patients with FLE was found to be significantly longer compared to HC (level 1: U = 2994.50, p = 0.000; level 2: U = 3101.00, p = 0.000; level 3: U = 2559.50, p = 0.000; level 4: U = 2184.50, p = 0.000). The total fixation duration on AOI/total completion time of FLE patients was significantly lower than those of HC for levels 1 to 3 (level 1: U = 1557.00, p = 0.000; level 2: U = 2333.00, p = 0.000; level 3: U = 2757.00, p = 0.000). Furthermore, the eye tracking data during the memory decoding phase were correlated with neuropsychological scale scores (p < 0.05).Conclusion Patients with FLE exhibited short-term memory impairment probably due to deficits in attentional maintenance, especially during the memory decoding phase. Eye tracking technology provided the possibility to help separate and quantify visual attention from memory processing, contributing to exploring underlying mechanisms of memory impairment in FLE.