High-speed inspection of inner diameter (ID) and outer diameter (OD) surface defects on thick-wall steel pipes is an important aspect to advance the pipeline inline inspection (ILI) in the oil and gas industry. The state-of-the-art ILI methods including magnetic flux leakage (MFL), ultrasonic testing (UT), electromagnetic acoustic transducer (EMAT), eddy current (EC) and pulsed eddy current (PEC) are hardly applicable to practical high-speed pipeline ILI due to the reasons of long inspection time required for sensor response and low detection sensitivity caused by severe motion. New sensing techniques that offer faster inspection speed, deeper signal penetration depth, better detection sensitivity and linearity, as well as capability of ID/OD discrimination are imperatively needed. This paper proposes a novel PEC sensing method to detect and discriminate ID/OD defects by utilizing the conductivity-dependent and permeability-dependent distribution patterns of induced eddy current at the ID surface of steel pipes. For ID defects, the pattern is caused by the discontinuous conductivity distribution, while for OD defects, the pattern is caused by the non-uniform incremental permeability distribution. A pulse-excited current with short width (2.5 μs), low duty cycle (1%) and fast-falling edge (100 ns) is injected into an excitation coil, so that the secondary magnetic field at the fast-falling edge will produce a transient oscillation in a pair of differential pick-up coils. Then, the time-domain transient oscillatory pick-up signal is extended, filtered, amplified, and extracted to be one feature by an envelope detector and an average sample method, which is processed in real-time by the developed probe for facilitating the back-end data recording. Meanwhile, a novel high-speed pipeline inspection gauge (PIG) with a sensor array is developed for field testing to validate the effectiveness of the proposed PEC method that achieved high inspection speed, deep detection depth, superior sensitivity, good linearity, low power consumption, easy implementation, ID/OD discrimination and crack detection capability.