Block transmission techniques have received much research interest recently for their ability to eliminate the inter-symbol interference problem associated with continuous transmission. Numerous detection schemes for single carrier and multi-carrier have been proposed. While these schemes may be derived from different principles, they usually rely on some prior estimate of the channel. However, channel estimation is usually affected by errors and most of existing detection schemes are known to be sensitive to such errors. In this paper, we develop a robust detection scheme that explicitly accounts for channel estimation error by optimizing the worst-case performance over a properly selected bounded uncertainty set. Although the prior channel estimation error is generally not bounded, we show that it is beneficial to refine the channel estimate over a properly chosen bounded uncertainty set of the channel centered on the prior channel estimate. Numerical results show an improved performance using the proposed robust approach over the one that ignores the prior estimation errors.