The continuous and noninvasive blood pressure (BP) measurement based on pulse transit time (PTT) doesn't need cuff and could monitor BP in real time for a long period. However, PTT is just a time index derived from electrocardiogram (ECG) and photoplethysmogram (PPG), while BP-related information within the PPG waveform has seldom been taken into consideration. We hypothesized that PPG waveform feature might be useful for BP estimation. Nine healthy subjects took part in an exercise stress test, including baseline resting, exercise on bicycle ergometry and recovering resting. ECG of lead V5 and PPG from left finger were collected simultaneously, and systolic blood pressure (SBP) and diastolic blood pressure (DBP) were recorded from a cuff sphygmometer on the right wrist. The correlation coefficients were obtained between BP (SBP, DBP and pulse pressure (PP)) and PPG morphological indices (total 15 indices in terms of waveform amplitude, time span and area ratio). Five PPG indices were correlated with both SBP and PP (absolute value of correlation coefficient |r| > 0.6) and were further tested for the capability to BP estimation, which were: (1) PTTA, time delay between the R peak of ECG and the foot point of PPG; (2) RSD, time ratio of systole to diastole; (3) RtArea, area ratio of systole to diastole; (4) TmBB, time span of PPG cycle; (5) TmCA, diastolic duration. Comparisons were made between the measured BP and the estimated BP by regression lines and quadratic curve fitting, respectively. As a result, the mean errors of SBP liner fitting with RSD, RtArea, TmBB and TmCA respectively were 5.5, 5.4, 5.2, 5.1 mmHg, which were smaller than that with PTTA of 5.8 mmHg. And the mean errors of SBP quadratic curve fitting with RSD, RtArea, TmBB and TmCA were all 5.1 mmHg, which were smaller than that with PTTA of 5.7 mmHg. The mean errors of multiple regression for SBP, PP and DBP was 4.7, 4.7, 3.5 mmHg respectively, which were more accurate than the regression with single PTTA of 5.8, 5.3, 5.2 mmHg respectively. However, PPG-based SBP and DBP could under estimate cuff pressure by 8 mmHg and over estimate by 10 mmHg respectively, which is a clinically significant error. In conclusion, the combination of time span (PTT, time ratio of systole to diastole, time span of PPG cycle and diastolic duration) and waveform morphology (area ratio of systole to diastole) could improve the performance of PPG-based BP estimation.