Systems that generate natural language text from abstract meaning\nrepresentations such as AMR are typically evaluated using automatic surface\nmatching metrics that compare the generated texts to reference texts from which\nthe input meaning representations were constructed. We show that besides\nwell-known issues from which such metrics suffer, an additional problem arises\nwhen applying these metrics for AMR-to-text evaluation, since an abstract\nmeaning representation allows for numerous surface realizations. In this work\nwe aim to alleviate these issues by proposing $\\mathcal{M}\\mathcal{F}_\\beta$, a\ndecomposable metric that builds on two pillars. The first is the principle of\nmeaning preservation $\\mathcal{M}$: it measures to what extent a given AMR can\nbe reconstructed from the generated sentence using SOTA AMR parsers and\napplying (fine-grained) AMR evaluation metrics to measure the distance between\nthe original and the reconstructed AMR. The second pillar builds on a principle\nof (grammatical) form $\\mathcal{F}$ that measures the linguistic quality of the\ngenerated text, which we implement using SOTA language models. In two extensive\npilot studies we show that fulfillment of both principles offers benefits for\nAMR-to-text evaluation, including explainability of scores. Since\n$\\mathcal{M}\\mathcal{F}_\\beta$ does not necessarily rely on gold AMRs, it may\nextend to other text generation tasks.\n