Methods of encoding and decoding sign information for non-zero coefficients in a block of video with transform skipped. The methods include context-based coding of a flag that signals if all signs are the same in the block; using a finite state machine to determine contexts for coding of sign information, some of which may be context-based coded and some of which may be non-context-based coded; context-based coding of sign information wherein the context is based upon the signs of neighbouring coefficients; context-based coding of sign information, where the sign information is a prediction of the sign; and using a dedicated context to code signs with a sign flag indicating whether signs in a transform skipping block have been inverted to preserve a dominant sign bias.