---
title: "Seven Process Modeling Guidelines (7PMG)"
type: source
tags: [bpm, modelling, quality, guidelines, empirical, bpmn, epc, error-probability, understandability]
authors: [Mendling, Jan; Reijers, Hajo A.; van der Aalst, Wil M.P.]
year: 2010
venue: "Information and Software Technology (Elsevier); preprint July 2009"
kind: paper
raw_path: "raw/Process Frameworks & BPM/2010-mendling-reijers-vanderaalst-7pmg.pdf"
sources: []
key_claims:
  - "7PMG synthesises seven concrete, empirically-grounded modelling guidelines (G1–G7) targeting model understandability and reduced error probability."
  - "G1: Use as few elements as possible — model size correlates positively with both error probability and difficulty of understanding (Mendling et al. studies on 600 SAP EPCs and 2000 industrial EPCs)."
  - "G2: Minimise routing paths per element — high connector degree (input + output arcs) correlates negatively with understanding and positively with errors."
  - "G3: Use one start and one end event — multiple starts/ends increase error probability and most workflow engines require single start/end."
  - "G4: Model as structured as possible — every split connector matched by a join of the same type (balanced brackets); unstructured models are both more error-prone and harder to understand."
  - "G5: Avoid OR routing elements — OR-join semantics carries paradoxes and implementation problems; AND/XOR-only models are less error-prone."
  - "G6: Use verb-object activity labels (e.g. 'Inform complainant') — empirically less ambiguous and more useful than action-noun ('Complaint analysis') or other styles (n=29 postgraduate experiment)."
  - "G7: Decompose model if it has more than 50 elements — error probability rises above 50% beyond this threshold; split sub-components with single-entry/single-exit into separate sub-process models."
  - "Empirical foundations rest on three measurement axes: process model understanding (questionnaire studies linking structural metrics to comprehension), error probability (relaxed soundness + EPC Soundness applied to industry corpora), and label ambiguity (controlled labeling-style experiments)."
  - "Guideline interactions are non-trivial: applying G2 (reduce degree) may increase model size in tension with G1; the paper offers a workshop-based industry priority ordering (21 modellers, German + Dutch practice) but no theoretical priority scheme."
  - "7PMG transformations preserve behaviour modulo branching bisimulation — the guidelines yield behaviour-equivalent but more understandable models, not behavioural changes."
created: 2026-05-04
updated: 2026-05-04
---

# Seven Process Modeling Guidelines (7PMG)

Mendling, Reijers and van der Aalst's synthesis of empirical process-modelling research into **seven actionable guidelines** for casual modellers. Distinguishes itself from prior frameworks (SEQUAL, Guidelines of Modeling) by being **operational** (each guideline is concrete and immediately applicable) and **empirically grounded** (each builds on quantitative studies).

## Motivating gap

Existing model-quality work splits into four streams the paper finds insufficient for practitioners:

1. **Top-down quality frameworks** (SEQUAL, GoM) — theoretically clean, distinguish quality categories, but **too abstract to be applied by novices**.
2. **Bottom-up metrics** (Cardoso CFC, Canfora maintainability, Mendling et al. error metrics) — fragmented, partial empirical validation.
3. **Empirical surveys of modelling techniques** — language-level conclusions (e.g., EPC vs Petri net understandability), not directly actionable on a single model.
4. **Pragmatic guidelines from practice** (e.g., "make your models hierarchical") — actionable but anecdotal, no empirical foundation.

7PMG sits in the middle: actionable like (4), empirically grounded like (2). Real-world modelling projects show error rates of 10–20% in industry corpora, motivating the need for simple guidance.

## Empirical foundations

The seven guidelines build on three measurement axes from prior Mendling-Reijers-Aalst studies:

- **Process model understanding** — questionnaire study (n=73 students across TU/e, Madeira, Vienna). Found OR-joins and average connector degree negatively correlated with comprehension.
- **Error probability** — prediction functions for relaxed soundness on the **SAP Reference Model (600 EPCs)** and EPC Soundness on **2000 industrial EPCs**. Trace error probability back to structural metrics; **size and complexity** are the dominant drivers.
- **Ambiguity of activity labels** — experiment with 29 TU/e postgraduates comparing verb-object ("send letter") vs. action-noun ("letter sending") vs. other styles. Verb-object significantly less ambiguous and more useful.

## The seven guidelines (Table 1)

| # | Guideline | Empirical basis |
|---|---|---|
| **G1** | Use as few elements in the model as possible | Larger models harder to understand and have higher error probability |
| **G2** | Minimise the routing paths per element | High connector degree (in-arcs + out-arcs) correlates with errors and reduced understanding |
| **G3** | Use one start and one end event | Multiple starts/ends increase error probability; workflow engines often require single start/end; enables soundness checks |
| **G4** | Model as structured as possible | Every split matched by join of same type — "balanced brackets". Unstructured models more error-prone, less understandable |
| **G5** | Avoid OR routing elements | Semantic paradoxes in OR-join, implementation difficulties; AND/XOR-only models less error-prone |
| **G6** | Use verb-object activity labels | Verb-object significantly less ambiguous and more useful than action-noun or rest-category labels |
| **G7** | Decompose a model with more than 50 elements | Error probability exceeds 50% above this threshold; split single-entry-single-exit subcomponents into sub-processes |

Each guideline preserves model behaviour modulo branching bisimulation when applied as a transformation — the resulting model is more understandable but **behaviour-equivalent**.

## Worked example (Section 3.4)

The paper uses a complaint-handling EPC from a Dutch governmental agency (37 elements, multiple start/end, OR-join, mixed labelling styles). Sequential application of the guidelines produces a transformed model with 31 elements, single start/single end, structured nesting, no OR-join, verb-object labels — same behaviour, demonstrably easier to read.

## Prioritisation (Section 4)

Guidelines can conflict: applying **G2** (reduce connector degree) may add elements, violating **G1**. The paper takes a pragmatic approach — workshops with 21 professional modellers (7 in Germany — Berliner BPM-Offensive; 14 in the Netherlands — major consultancy) — to elicit priority ordering. *Theoretically motivated, empirically validated prioritisation is acknowledged as out of scope for this paper.*

## Limitations & merits (Section 5)

- Limited to **control-flow perspective** — does not cover data, organisation, or compliance dimensions.
- Some guidelines (e.g., G1, G7) are **size-based heuristics**, not formal correctness criteria — pair with verification (cf. soundness, [[sources/1998-vanderaalst-verification-of-workflow-nets]]) for full quality coverage.
- Guidelines are **language-agnostic** in spirit (illustrated on EPCs, applicable to BPMN, YAWL, UML AD) but the empirical studies were predominantly EPC-based.
- Builds on **branching-bisimulation equivalence** — guidelines describe transformations within a behaviour-equivalence class, not redesigns.

## Connections

**Concepts:** [[concepts/process-model-quality]] · [[concepts/7pmg]] · [[concepts/process-model-complexity-metrics]] · [[concepts/soundness]] (G3 enables soundness analysis)
**Frameworks:** [[frameworks/bpmn]] (canonical target notation today) · EPCs (paper's primary illustration)
**Related sources:** [[sources/2018-dumas-fundamentals-of-bpm]] (§5.4.4 cites 7PMG) · [[sources/1998-vanderaalst-verification-of-workflow-nets]] (formal soundness — complementary verification layer) · [[sources/2006-krogstie-sindre-jorgensen-revised-sequal-framework]] (the SEQUAL framework 7PMG positions itself against)
**Authors / hub entities:** [[entities/jan-mendling]] · [[entities/hajo-reijers]] · [[entities/wil-van-der-aalst]]

## Open follow-ups
- Theoretically-motivated prioritisation scheme (acknowledged gap).
- Extension beyond control-flow: data-quality, organisational-quality guidelines.
- Empirical re-validation on BPMN corpora (original studies were EPC-heavy).