We present a fully abstract encoding of λ ref , the call-by-value λ-calculus with references, in the πcalculus. By contrast with previous full abstraction results for sequential languages in the π-calculus, the characterisation of contextual equivalence in the source language uses a labelled bisimilarity. To define the latter equivalence, we refine existing notions of typed bisimulation in the π-calculus, and introduce in particular a specific component to handle divergences. We obtain a proof technique that allows us to prove equivalences between λ ref programs via the encoding. The resulting proofs correspond closely to normal form bisimulations in the λ-calculus, making proofs in the π-calculus expressible as if reasoning in λ ref. We study how standard and new up-to techniques can be used to reason about diverging terms and simplify proofs of equivalence using the bisimulation we introduce. This shows how the π-calculus theory can be used to prove interesting equivalences between λ ref programs, using algebraic reasoning and up-to techniques.