BACKGROUND: Non-convulsive seizure (NCS) is an underdiagnosed, potentially treatable emergency with significant mortality and morbidity. The objective of this study is to examine the characteristics of patients with NCS presenting with altered mental status (AMS) and diagnosed with electroencephalography (EEG), to identify the factors that could increase the pre-test probability of NCS. METHODS: Retrospective study using the data collected prospectively. Inclusion criteria:patients older than 13 years with AMS. Exclusion criteria: (1) immediately correctable AMS (e.g., hypoglycemia, opiod overdose); (2) inability to undergo EEG; (3) hemodynamic instability. Outcomes compared between NCS and non-NCS cases: vital signs, lactate level, gender, witnessed seizure, use of anticonvulsive in the field or in the ED, history of seizure or stroke, head injury, abnormal neurological finding and new abnormal findings on head CT. Data presented as medians and quartiles for categorical and percentages with 95％CI for continuous variables. Univariate analyses were performed with Man-Whitney U and Fisher's Exact tests. A multivariate analysis model was used to test the predictive value of clinical variables in identifying NCS. RESULTS: From 332 patients (median age 66 years, quartiles 50–78), 16 were diagnosed with NCS (5％, 95％CI 3％–8％). Only age was significantly different between the NCS vs. non-NCS groups in both univariate (P=0.032) and multivariate analyses (P=0.016). CONCLUSION: Other than age, no other clinically useful variable could identify patients at high risk of NCS. ED physicians should have a high suspicion for NCS and should order EEG for these patients more liberally.