The plastid trnL-F region has proved useful in molecular phylogenetic studies addressing diverse evolutionary questions from biogeographic history to character evolution in a broad range of plant groups. An important assumption for phylogenetic reconstruction is that data used in combined analyses contain the same phylogenetic signal. The trnL-F region is often used in combined analyses of multiple chloroplast markers. These markers are assumed to contain congruent phylogenetic signal due to lack of recombination. Here we show that trnL-F sequences display a phylogenetic signal conflicting with that of other chloroplast markers in Annonaceae, and we demonstrate that this conflict results from ancient paralogy. TrnL-F copy 2 diverged from trnL-F copy 1 (as used in family-wide phylogenetic analyses) in a direct ancestor of the Annonaceae. Although this divergence dates back 88 million years or more, the exons of both copies appear to be intact. In this case, assuming that (putative) chloroplast markers contain the same phylogenetic signal results in an incorrect topology and an incorrect estimate of ages. Our study demonstrates that researchers should be cautious when interpreting gene phylogenies, irrespective of the genome from which they are presumed to have been sampled.