The parameter identification of a six-axis accelerometer is a difficult problem due to its higher input and output volumes and linear dynamic equation. According to this, a four-step method was proposed to identify the 25 decoupling parameters of a parallel type six-axis accelerometer. A calibration platform based on double slider-crank mechanisms was designed and processed to provide the external stimulation and a virtual instrument based on LabVIEW was developed to provide the software support for the parameter identification. The first sets of parameters were identified by averaging pretreatment data in static state. By putting the sensor on the platform to do a pure line movement with the frequency of 1-2 Hz, the dynamic equations were simplified to linear algebraic equations, then the second set of parameters were identified by using the least square method. Similarly, the third set of parameters were identified when the sensor did a pure angular movement with the frequency of 1-2 Hz. The fourth set of parameters were identified by one-dimensional searching about stiffness to mass ratio when the sensor did the pure line movement with the frequency of 4-5 Hz. Experimental results indicate that the maximum relative error is 7.479% after using the identified parameters to decouple the six-axis acceleration, which reduces a magnitude compared to that before parameter identification. Above results verify that the proposed four-step method is correct and feasible.