Wide-Area Monitoring Systems (WAMS) have been widely accepted to provide real-time monitoring, protection, and control of power systems. The huge and rapidly increasing data volume in WAMS imposes a heavy burden on the communication and storage systems and could become the bottleneck for many real-time smart grid applications. This paper presents a framework for intelligent lossy compression in a real-time manner for synchrophasor data in WAMS. The proposed method is capable of achieving good compression ratios without introducing impractical delays and sacrificing much accuracy. Because disturbance data has much stricter delay and fidelity requirements, an early disturbance detection technique is introduced to identify disturbance data and handle it differently. The performance of the method is demonstrated by experiment results on real synchrophasor data collected from multiple substations.