While it is generally accepted that temperature determines the approximate altitude of treeline, drivers of fine-scale treeline dynamics are poorly understood. A dominant driver of fine-scale dynamics is sheltering of seedling by adult trees. Within the treeline literature, there is an underlying assumption that sheltering from trees is beneficial for seedlings, initiating a positive feedback loop. However, sheltering can potentially create both hospitable and hostile microclimates. In this study, we examined the relationship between microclimate and seedling distribution at an abrupt treeline on Pikes Peak, Colorado and ask two main questions: (1) how do trees modify their surrounding microclimate, and (2) how do these microclimates impact seedling establishment? To explore these questions, we analyzed 2017 daytime and nighttime thermal images of treeline, and 2017 seedling data mapped with 20 cm precision. We first analyzed the spatial distribution of surface temperature with respect to the treeline spatial structure, followed by seedling density with respect to ground temperatures. Overall, we found that variations within the microclimates surrounding adult trees are determined by three primary factors: (1) the obstruction of boundary layer airflow, which results in the formation of eddy-like isolated pockets of air that exhibit exaggerated diurnal temperatures (2) long wave radiation from trees, a nighttime phenomenon that warms a certain radius of the ground beneath a tree, and (3) shade, which accounts for the majority of low daytime temperatures. These three competing mechanisms interact, resulting in a mosaic of variable ground temperatures adjacent to trees. Between these variable microclimates, seedlings exhibited distinct temperature preferences, with the highest density at both moderate daytime and nighttime temperatures, and an explicit avoidance of both daytime and nighttime high and low temperatures. When temperature is divided into 12 equal classes and seedling density is compared between the classes, the best fit for both the daytime and nighttime datasets were third degree polynomials (Daytime: adjusted R2= 0.70, F(3,8)= 9.458, p=0.005, Nighttime: adjusted R2=0.58, F(3,8)=6.043, p=0.019). An analysis using a relative distribution estimate function (Rhohat) in spatstats package in R corroborated these findings.