In contrast to the customary approximations based on standard normal percentage points, direct approximations involve simple functions of parameters (such as degrees of freedom and tail area of the t distributions). This article used techniques of exploratory data analysis following Hoaglin to develop direct approximations for percentage points in the commonly used portions of upper tail of the t distribution with small to moderate numbers of degrees of freedom. These approximations are convenient to use and they compare favorably in accuracy with the popular approximations based on standard normal percentage points such as Peiser"s. They can be used as an initial value generator in algorithms for getting more accurate percentage points.