The efficiency of the multilevel fast multipole algorithm (MLFMA), which needs the interpolation and anterpolation (I&A) processes between consecutive levels, is improved by using the isosceles triangular interpolation (TI) instead of the commonly used Lagrange interpolation (LI). For the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$p$ </tex-math></inline-formula> th-order interpolation where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$p >0$ </tex-math></inline-formula> , the number of sampling data points used for the TI is ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$p +1$ </tex-math></inline-formula> )( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$p +2$ </tex-math></inline-formula> )/2, much less than the LI, which needs ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$p +1$ </tex-math></inline-formula> )2 points. To effectively use the TI, the sampling points on the Ewald sphere should be reselected to form isosceles triangular data grids but not the common rectangular ones. However, the numerical integration in the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\theta $ </tex-math></inline-formula> -direction over the Ewald sphere is based on the Gauss–Legendre quadrature rule, resulting in nonuniform triangular grids. A novel TI approach suitable for the nonuniform grids is proposed, while the interpolation error in the TI is analyzed in detail. Numerical results of typical electromagnetic (EM) scattering cases are shown to illustrate the accuracy and efficiency of the proposed TI-based MLFMA.