Abstract. We study the Kolmogorov-Petrovsky-Piskunov minimal front speeds in spatially random shear flows in cylinders of various cross sections based on the variational principle and an associated elliptic eigenvalue problem. We compare a standard finite element method and a two-scale finite element method in random front speed computations. The two-scale method iterates solutions between coarse and fine meshes and reduces the cost of the eigenvalue computation to that of a boundary value problem while maintaining the accuracy. The two-scale method saves computing time and provides accurate enough solutions. In the case of square and elliptical cross sections, our simulation shows that larger aspect ratios of domain cross sections increase the average front speeds in agreement with an asymptotic theory. 1. Introduction. Front propagation in heterogeneous flows is an active research area in applied science and mathematics [7,14,15,18,19,25,32,33,36]. A fundamental problem is to characterize and compute large-scale front speeds in random flows [10,19,30,31,39]. The Kolmogorov-Petrovsky-Piskunov (KPP) minimal front speeds admit a variational characterization in terms of the principal eigenvalue or principal Lyapunov exponent of an associated linear operator [6,5,16,37,28,29]. The variational principle of KPP front speeds makes accurate and efficient analytical and numerical studies possible. It is known that KPP front speeds are enhanced by spatially random shear flows in cylinders, with a quadratic (linear) law in the small (large) root mean square amplitude regime; see [28] and the references therein. However, less is known about how the domains influence the front speeds.In this paper, we shall study the dependence of KPP front speeds (in spatially random shear flows) on the aspect ratio of cylindrical cross sections. We shall use both numerical methods and asymptotic methods. Let D ≡ R × Ω, where Ω ⊂ R