AbstractTo enhance the activity of transketolase towards nonphosphorylated substrates and enlarge the scope of its substrates, notably to long polyol aldehyde acceptors (D‐ribose or D‐glucose), a rational design‐supported evolution strategy was applied. By using docking experiments, an in silico library, and iterative mutagenesis, libraries of single‐ and double‐point mutants were designed and generated. A double‐screening approach was implemented, coupling a preselection activity assay (HPLC method) and a selective assay (GC method) to find the best enzymes. Several mutants (R526N, R526Q, R526Q/S525T, R526K/S525T) showed improved activities towards nonphosphorylated substrates as the coupled products of lithium hydroxypyruvate (HPA) with glycolaldehyde (GO), D‐ribose or D‐glucose. These mutated enzymes were further characterised. They were shown to be up to four times more active than the wild‐type (mutant R526Q/S525T) for nonphosphorylated substrates LiHPA/GO (Vm/Km for LiHPA = 92.4 instead of 28.8×10−3 min−1 for the wild‐type) and 2.6 times more active for substrates LiHPA/rib.