Abstract: A novel time domain neural networks based structural stiffness and damping parameters identification methodology with the direct use of structural acceleration time series is proposed, which is called direct soft parametric identification (DSPI). The theoretical fundamentals of the methodology is explained and the architecture of the two neural networks is described according to the discrete time solution of the state space equation of the structural vibration differential equation. An evaluation index called the root mean square of the acceleration prediction difference vector (RMSAPDV) is defined and employed to identify structural parameters. Based on an acceleration-based neural network modeling (ANNM) for the reference structure which parameters are determined by an estimation of an object structure, and a parameter evaluation neural network(PENN) that describes the relation between structural parameters and the components of the corresponding RMSAPDVs, the inter-storey stiffness and damping ratio of a frame model structure excited by a shaking table with known mass distribution, are identified by the direct use of acceleration measurements. Compared with the identified structural parameters from sweep frequency tests, results show that structural parameters can be identified with acceptable accuracy and the proposed method is an applicable approach closing to real-time direct identifications.