Research on emotion states based on physiological signals has attracted more and more attention. However, existing studies mainly focused on the classification accuracy tests based on different machine learning methods, without a comprehensive exploration for the inherent links between physiological features and emotion states. This pilot study aimed to investigate the differences of heart rate variability (HRV) indices between two opposite emotional states: happiness and sadness, to reveal the differences of autonomic nervous system activity under different emotional states. Forty-eight healthy volunteers were enrolled in the proposed study. Electrocardiography (ECG) signals were recorded under two emotion states with a random measurement order (first happiness then sadness or reverse). RR interval (RRI) time series were extracted from ECGs and multiple HRV indices, including time-domain (MEAN, SDNN, RMSSD and PNN50), frequency-domain (LFn, HFn and LF/HF) and nonlinear indices (SampEn and FuzzyMEn) were calculated. The results showed that experimental order had no significant impact on all HRV indices. Among all nine indices, six indices were identified having significant differences between happiness and sadness emotion states: MEAN (P<0.05), SDNN (P<0.01), three frequency-domain indices (all P<0.01) and FuzzyMEn (P<0.05), whereas RMSSD, PNN50 and SampEn were reported having no significant differences among the two emotional states. All nine indices except for SampEn had significant positive correlations (all P<0.01) for the two emotion states. All indices showed no HR-related or MAP-related changes for each emotional state except that four time-domain indices decreased with the increase of HR (P<0.01). It concluded that HRV indices had significant differences between happiness and sadness emotion states and the findings could help to better understand the inherent differences of cardiovascular time series between different emotion states in clinical practice.