Abstract: Based on the official forecasts of tropical cyclone (referred as TC hereinafter) in the Northwest Pacific from 1999 to 2003, real time error analysis is carried out for China Central Meteorological Observatory (CMO), Japan Meteorological Agency (JMA) and Joint Typhoon Warning Center (JTWC). Five aspects such as total error, annual error, different sea-areas, different tracks and different intensity trends are considered to compare the error characteristics of each forecast center in detail. Evidence suggests that JTWC's path error is the least and JMA has the best behavior in intensity forecast from a view of 5-year average. In great probability, all centers may overestimate TC's intensity, but when underestimating occurs, the bias always turn out to be a bigger one. After the year of 2000, all centers' performances improve greatly. As for CMO, Its track errors are less than 150 km (24 h) and 250 km (48 h) respectively. Over different see-areas, there is always a better path forecast among the three centers:JTWC behaves better at West-Luzon area. JMA has a overwhelming superiority on others at Che ju area. Its 24 h error is just 88 km, while the others produce an average of about 130 km. At South-Ryukyu area, JMA and JTWC perform obviously better than CMO whose track bias averages out to 155 km in 24 h and 340 km in 48 h forecast, while others just make out a value of 110 km and 280 km respectively. In most cases, JMA's intensity forecast is the best, but there isn't a obvious gap between CMO and JMA. JTWC's result is the poorest. As to different TC tracks, except for those veering at South China Sea, JTWC's path forecast is often the best. For intensity forecast, all centers make poor behaviors in turnabout track, while JMA and CMO perform better in Veering-after-landing and Northeastward tracks. For tropical cyclones with different intensity trends, JTWC mostly gives out a better path, especially for rapid intensification cases. Generally, all members make poorer path forecast for weakening cases. Intensity error of steady TC is the smallest, while that of rapid intensification is particularly bigger than others. That means more attention should be paid on researching and forecasting of rapid change TC, including rapid intensifying and rapid weakening.These results help to understand entirely forecast qualities of the three main forecast centers and give some guidance for forecasters when they make references on these centers' forecast results, and above statistics and analysis can be either a base or a direction for concerning research.