From the magnetoelastic effect principle, it is known that diversity exists in the hysteresis loops of rodelike ferromagnetic material under different tension. So it can be employed to develop an improved method for tension estimation of rodelike structure. Firstly, the hysteresis loop signals of rodlike structure are acquired by an EM sensor which is composed of two coaxial solenoid coils. Then, a curve of hysteresis loop change (CHLC) is defined to reflect the tension influence on each point of the hysteresis loop. Secondly, lowresolution CHLC is obtained by using wavelet analysis. Finally, neural network is used to establish the relationship between the lowresolution CHLC and tension after the data with different tension is used to train the neural network. The experimental results show that CHLC can reflect the tension influence on each point of hysteresis loop intuitively. The lowresolution CHLC not only has the characteristics of including the entire information of tension but also has the characteristics of low dimensionality. The relationship between the lowresolution CHLC and tension can be obtained by using neural network, without analysis of the sensitivity curve and coefficient of determination curve. The simple linear interpolation based RBF neural network has a better performance than BP neural network and RBF neural network. The tension measurement method based on lowresolution CHLC and simple linear interpolation based RBF neural network is applied to two coaxial solenoid coils based EM sensor to measure the tension and its average prediction error and coefficient of determination are 0.11% and 1, respectively. This method is effective and can meet the actual measurement requirements.