Objective: High frequency oscillations (HFO) in scalp EEG are a new and promising epilepsy biomarker. However, considerable fluctuations of HFO rates have been observed through sleep stages and cycles. Here, we aimed to identify the optimal timing within sleep and the minimal data length for sensitive and reproducible HFO detection. Methods: We selected 16 whole-night scalp EEG recordings of paediatric patients with a focal structural epilepsy. We used an automated clinically validated HFO detector to determine HFO rates (80–250 Hz). We evaluated the reproducibility of HFO detection across intervals. Results: HFO rates were higher in N3 than in N2 and REM (rapid eye movement) sleep and highest in the first sleep cycle, decreasing with time in sleep. In N3 sleep, the median reliability of HFO detection increased from 67% (interquartile range: iqr 57) to 78% (iqr 59) to 100% (iqr 70%) for 5-, 10-, and 15-min data intervals, improving significantly (p = 0.004, z = 2.9) from 5 to 10 min but not from 10 to 15 min. Conclusions: We identified the first N3 sleep stage as the most sensitive time window for HFO rate detection. At least 10 min N3 data intervals are required and sufficient for reliable measurements of HFO rates. Significance: Our study provides a robust and reliable framework for scalp HFO detection that may facilitate their implementation as an EEG biomarker in paediatric epilepsy.