Various training methods for two way memory effect (TWME) and stressed two way memory effect (STWME) were tried on Cu 13.4Al 4.0Ni (mass fraction, %) single crystals by applying tensile stress along 〈001〉 direction of β phase. The training method of cooling with load can induce a lot of martensite prone to stabilize, thus cause large residual deformation, wide hysteresis and small TWME. Training with constant load can produce STWME larger than 8% with the least residual deformation. By training procedure of martensite reorientation below M f followed by thermal cycling, the TWME is relatively large with very small residual deformation and with comparatively narrow hysteresis of two way memory. The obtained two way memory curve after such training is not a closed loop, and the obtained TWME is not stable. However, these can be improved by thermal cycling. Training with martensite reorientation below M f and thermal cycling under relatively low constant stress throughout the whole training procedure is the optimum way of obtaining TWME, and more than 1.7% TWME can be obtained. The thermomechanical history of the sample has a pronounced effect on the training result. Thermomechanical cycling has a softening effect on martensite. [ Various training methods for two way memory effect (TWME) and stressed two way memory effect (STWME) were tried on Cu 13.4Al 4.0Ni (mass fraction,%) single crystals by applying tensile stress along <001> direction of beta phase. training method of cooling with load can induce a lot of martensite prone to stabilize, thus cause large residual deformation, wide hysteresis and small TWME. Training with constant load can produce STWME larger than 8% with the least residual deformation. By training procedure of martensite reorientation below M f followed by thermal cycling, the TWME is relatively large with very small residual deformation and with comparatively narrow hysteresis of two way memory. The obtained two way memory curve after such training is not a closed loop, and the resulting TWME is not stable. However, these can be improved by thermal cycling. Training with martensite reorientation below M f and thermal cycling under relatively low constant stress thro ughout the whole training procedure is the optimum way of obtaining TWME, and more than 1.7% TWME can be obtained. The thermomechanical history of the sample has a pronounced effect on the training result. Thermomechanical cycling has a softening effect on martensite. [