Most traditional theories of speech production are currently based on plane waves and on one-dimensional analysis. It i s however well-known that when the frequency of sound reaches a cut-on frequency, higher acoustical modes start to propagate and can become predominant. It is therefore important t o evaluate the effects of these higher modes, especially in order t o improve acoustical models of the vocal tract. This paper describes a new numerical method to study the propagation and the radiation of speech sounds, and to compute acoustic characteristics of the vocal tract. This method, named Transmission Line Matrix or Modelling (TLM), has been used for simulating electromagnetic wave propagation and is used here for the first time in acoustics. The TLM method provides time domain solutions in 2D and 3D spaces. The main advantage of this method is the simplicity of formulation and programming for a large range of applications. We first describe the principle on which the TLM method is based. The method as well as the boundary conditions used are validated using classical tests. A systematic study of higher order mode propagation and radiation is then presented. We focus on the influence of some critical parameters such as vocal tract width and location of the sound source. In particular, we show how, using TLM simulation, it is possible to derive modal reflection and transmission characteristics of the vocal tract. A typical example of simulation is presented and discussed.