With the rise of the popularity and usage of neural networks, trustworthy uncertainty estimation is becoming increasingly essential. In this paper we present a computationally cheap extension of Deep Ensembles for a regression setting called Bootstrapped Deep Ensembles that explicitly takes the effect of finite data into account using a modified version of the parametric bootstrap. We demonstrate through a simulation study that our method has comparable or better prediction intervals and superior confidence intervals compared to Deep Ensembles and other state-of-the-art methods. As an added bonus, our method is better capable of detecting overfitting than standard Deep Ensembles.