The classical results of Lorenz indicate that regardless of the number and quality of observations, flows with many scales of motion may have finite limits of predictability, owing the increase of error growth rates as scale decreases. Turbulence closure models suggest a more subtle situation: predictability appears to be intrinsically limited in three-dimensional turbulence, but not in two-dimensional turbulence. More recently, increased computational capabilities have allowed direct investigation of the growth of forecast differences in flows that truly span multiple scales of motion. I will review recent efforts to quantify and understand error growth in simulations of baroclinic waves with explicitly resolved moist convection and in high-resolution simulations of quasigestrophic turbulence.