This paper presents a new method called phase of transient response using the local reference pixel vector (PTR-LRPV) to process pulsed thermography data for defect detection and depth estimation in carbon fiber reinforced polymer specimens. Due to the use of flash as excitation source in pulsed thermography and subsequently its adaptation with the conditions of suddenly applied input, the received signal from the infrared camera can be separated into two transient and steady-state responses in the frequency domain. Defects cause local variations in the thermal spatio-temporal patterns that the transient response can fairly reveal such highly informative variations. On the other hand, the steady-state response mainly includes the intrinsic characteristics of the specimen (sound areas). In fact, by properly separating these responses and employing the phase of the transient term, a suitable distinction of defect characteristics from sound areas has been reached. The results show that the proposed PTR-LRPV is effective in both defect detection and depth estimation tasks, and also, can fairly compete with several well-known algorithms in terms of both quantitative and qualitative criteria.