Despite decades of intense genetic, biochemical, and evolutionary characterizations of bacterial promoters, we still lack the basic ability to identify or predict transcriptional activities of promoters using primary sequence. Even in simple, well-characterized organisms such as E. coli there is little agreement on the number, location, and strength of promoters. Here, we use a genomically-encoded massively parallel reporter assay to perform the first full characterization of autonomous promoter activity across the E. coli genome. We measure promoter activity of >300,000 sequences spanning the entire genome and precisely map 2,228 promoters active in rich media. We show that antisense promoters have a profound effect on global transcription and how codon usage has adapted to encode intragenic promoters. Furthermore, we perform a scanning mutagenesis of 2,057 promoters to uncover regulatory sequences responsible for regulating promoter activity. Finally, we show that despite these large datasets and modern machine learning algorithms, the task of predicting promoter activity from primary sequence sequence is still challenging.