When selecting a digital oscilloscope, bandwidth is not a standalone number; the real-time sample rate must be sized to it, and the standard operating window is 3 to 5 samples per Hz of analog bandwidth for Gaussian-response front ends [S3][S5]. Tektronix publishes a "5 Times Rule" that ties bandwidth to amplitude accuracy: a scope running at 5x the highest signal frequency holds amplitude error under ±2% [S1].
Rohde & Schwarz tightens the lower floor to 2.5x: a 1.5 GHz scope needs at least 3.75 GS/s in real-time single-shot mode, with extra headroom recommended for transients and eye diagrams [S2]. National Instruments repeats the 3 to 4x rule in its 2026-09-03 oscilloscope selection guide, confirming the 3x minimum has become the de-facto specification baseline across the major T&M vendors [S7].
Bandwidth: the analog ceiling of the front end
Bandwidth on a modern digital scope is defined as the -3 dB point of the front-end amplifier, the frequency at which a sine input is attenuated to 70.7% of true amplitude, or roughly a 30% drop in displayed voltage [S1][S2]. Below that point the scope acts as a low-pass filter passing DC content cleanly; above it, edges blur, amplitude droops, and high-frequency harmonics roll off. The R&S buyer's guide gives the worked example: a 100 MHz sinewave is measured accurately with a 150 MHz scope, while a 133 MHz memory bus with 600 ps rise time carries frequency content up to 583 MHz, so the same signal needs a 400-700 MHz scope under the 3 to 5x rule [S2].
The 0.35 / rise-time formula gives the first-order bandwidth estimate for digital edges and is the same number embedded in vendor datasheets: a 1.75 ns rise time corresponds to roughly 200 MHz of analog content, matching the typical 200 MHz mid-range scope class [S2][S4]. Pushing past required bandwidth is not free; excess analog bandwidth also lifts broadband noise into the trace, which is why R&S ships a 20 MHz hardware filter and HD (high-definition) modes on MXO 4 and RTO 6 to trade ADC resolution against bandwidth [S2].
Sample rate: the digitizer's capture density
Real-time sample rate is the ADC's continuous GS/s, distinct from equivalent-time sampling which only applies to repeating signals. Keysight's reference puts the engineering rule at 3 to 5x the bandwidth, framing the lower bound as "high-enough" for transient capture and the upper end as the point at which interpolation and edge fidelity stop improving [S5]. All About Circuits narrows the same idea to the Gaussian-response case: 4 to 5x bandwidth is the usual real-time target because Gaussian roll-off attenuates alias-prone energy before the Nyquist limit, unlike a brick-wall response which keeps nearly all energy up to fS/2 [S3].
For a 200 MHz scope the typical floor is 1 GS/s on two channels interleaved to 2 GS/s on one channel, a ratio common in the Keysight 2000X / 3000X families discussed in the EEVblog thread [S4]. Two to three points on a 1.75 ns edge, sampled at 0.5 ns intervals, is what turns a square wave from a jagged staircase into a recognisable trapezoid; drop below one sample per rise time and the trigger logic can fire on ghosts of the edge instead of the edge itself [S4].
The bandwidth-to-sample-rate decision matrix

The four working regimes practitioners actually use, taken from the cross-vendor guidance, line up as follows: (1) 2.5x is the R&S absolute floor and is only acceptable for repetitive signals on a clean sine; (2) 3 to 4x is the NI minimum and covers most general-purpose debug on Gaussian-response mid-range scopes [S2][S7]; (3) 4 to 5x is the All About Circuits / Keysight target for transient capture and one-shot events where aliasing of out-of-band energy would otherwise fold back into the passband [S3][S5]; (4) anything above 5x is diminishing return unless the application is serial-bus eye masking or wideband RF, where the extra samples feed post-acquisition signal processing and de-embedding [S2][S4].
Cross-checked against cost: a 200 MHz / 1 GS/s bench scope sits in the entry tier, a 1 GHz / 5 GS/s mid-tier, and a 4 GHz / 20 GS/s high-end serial-analysis tier. The 2.5x, 5x and 10x multiples each map to a different vendor product class, and the data logger companion page on Nyquist-driven sample rate picks up the same algebra for slow signals. A direct comparison of brick-wall vs Gaussian response further explains why the same bandwidth-to-sample-rate ratio behaves differently across scope families.
Selection workflow: pick bandwidth, then size the ADC
Step 1: identify the highest-frequency content in the DUT. For sine sources, use the carrier. For digital buses, use 0.35 / rise-time first, then sanity-check against 3 to 5x the clock or data rate [S2]. Step 2: multiply that frequency by 5 if amplitude accuracy under ±2% matters, or by 3 if the signal is roughly sinusoidal and a 30% roll-off at the edge of band is acceptable [S1][S2]. Step 3: pick a scope whose rated real-time sample rate on the channels you will actually use is at least 4 to 5x the bandwidth you just selected, so the ADC is not the bottleneck [S3][S5].
Step 4: validate the chain at the probe tip, not the BNC. A 500 MHz scope paired with a 100 MHz passive probe measures a 100 MHz signal, and no amount of ADC headroom fixes that [S1]. Step 5: confirm channel-count de-rating. Many mid-range scopes halve the sample rate when all four channels are active because of ADC interleaving, so the spec sheet GS/s number is only valid under the channel configuration you intend to run [S4].
When the rules break: overshoot, undershoot, and edge cases

Three failure modes recur in field debugging. First, the "enough bandwidth, starved ADC" case: a 1 GHz scope running at 2.5 GS/s in 4-channel mode drops the effective ratio to 2.5x, violating the All About Circuits Gaussian floor and producing interleave spurs visible on FFT screens [S2][S3]. Second, the "excess bandwidth, noisy trace" case: a 2 GHz scope on a 50 MHz power-rail measurement displays 4x the noise floor; engage the 20 MHz hardware limiter or HD mode to recover vertical resolution, trading analog bandwidth for effective ENOB [S2].
Third, the "repetitive signal" trap: equivalent-time sampling can rebuild a 1 GHz eye diagram on a 200 MS/s ADC, but only if the signal is stable and triggerable, conditions rarely met during one-shot transient capture or in-circuit debugging with probe loading [S2][S4]. The eevblog forum discussion of 8-cascaded LMH6642 op-amps as a poor-man's anti-alias filter illustrates why scope vendors ship integrated front ends: hand-rolled filters ripple in the passband and drift with temperature, and the few pence saved per ADC do not justify a worse noise and step-response spec on a competitive bench instrument [S4].
Standards and sourcing notes for spec sheets
No IEC or ISO standard governs oscilloscope bandwidth or sample-rate ratios directly; vendors publish their own methodologies, and the 3x, 4x, 5x rules are industry conventions, not regulated values [S1][S2][S5]. Tektronix's "5 Times Rule" is sourced to its own primer, R&S's 2.5x floor is a buyer's-guide statement, and Keysight's 3 to 5x framing appears in its KKB knowledge base [S1][S2][S5]. The IEEE 1057 digitizing waveform recorder standard covers ADC testing methods, including effective-bit and alias-rejection tests, but does not prescribe a sample-rate-to-bandwidth ratio for general-purpose scopes.
For procurement, the practical signal is the per-channel real-time GS/s, not the interleaved marketing maximum, and the analog bandwidth specified at the BNC, not the probe tip. The 5x-bandwidth rule for amplitude, the 4 to 5x-bandwidth sample rate for Gaussian fronts, and the 0.35 / rise-time edge content formula are the three numbers an engineer should see on the datasheet before a scope is approved onto a lab bench.